{
  "generated_at": "2026-11-19T14:37:39Z",
  "demo": false,
  "hardware": {
    "gpus": [
      "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)"
    ],
    "backends": [
      "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 4 parallel slots)",
      "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "Vulkan (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots)",
      "Vulkan (Ollama 0.40.4, flash attention, 16k context; 2 parallel slots)",
      "Vulkan (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)"
    ]
  },
  "metrics": {
    "decode_tps": {
      "label": "Decode speed, one stream",
      "unit": "tok/s"
    },
    "decode_tps_c4": {
      "label": "Decode speed, 4 parallel streams (total)",
      "unit": "tok/s"
    },
    "ttft_s": {
      "label": "Time to first token",
      "unit": "s"
    },
    "prefill_tps_4k": {
      "label": "Prompt processing, about 4k tokens",
      "unit": "tok/s"
    },
    "prefill_tps_12k": {
      "label": "Prompt processing, about 12k tokens",
      "unit": "tok/s"
    },
    "load_seconds": {
      "label": "Load time after a fresh unload",
      "unit": "s"
    },
    "size_gb": {
      "label": "Model size as loaded on our server",
      "unit": "GB"
    },
    "vram_gb": {
      "label": "GPU memory on our server",
      "unit": "GB"
    },
    "gpu_fraction": {
      "label": "Share of the model on the GPU",
      "unit": "ratio"
    },
    "context_length": {
      "label": "Context length loaded",
      "unit": "tokens"
    }
  },
  "note": "Measured on one laptop with Ollama; speeds are comparable only between models run on the same backend.",
  "count": 236,
  "models": [
    {
      "slug": "qwen3-0-6b",
      "display_name": "Qwen3 0.6B",
      "model_version_id": 165,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 0.6,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 43.9956,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 222.9919,
        "decode_tps_c4": 261.1199,
        "ttft_s": 0.0203,
        "prefill_tps_4k": 8611.2121,
        "prefill_tps_12k": 5013.0401,
        "load_seconds": 66.126,
        "size_gb": 15.6378,
        "vram_gb": 15.6378,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "exaone3-5-2-4b-instruct",
      "display_name": "Exaone3.5 2.4B Instruct",
      "model_version_id": 94,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.4,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 76.256,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 100.8843,
        "decode_tps_c4": 256.0,
        "ttft_s": 0.0402,
        "prefill_tps_4k": 2889.9137,
        "prefill_tps_12k": 2324.5215,
        "load_seconds": 74.297,
        "size_gb": 11.8868,
        "vram_gb": 11.8868,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-moe-3b-instruct",
      "display_name": "Granite3 Moe 3B Instruct",
      "model_version_id": 121,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 47.72,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 147.4023,
        "decode_tps_c4": 218.43,
        "ttft_s": 0.0373,
        "prefill_tps_4k": 3846.3786,
        "prefill_tps_12k": 4339.7265,
        "load_seconds": 11.375,
        "size_gb": 4.2969,
        "vram_gb": 4.2969,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "smollm2-135m-instruct",
      "display_name": "Smollm2 135M Instruct",
      "model_version_id": 175,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 0.135,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 20.1171,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 179.8502,
        "decode_tps_c4": 201.1696,
        "ttft_s": 0.025,
        "prefill_tps_4k": 4627.7033,
        "prefill_tps_12k": 4923.8474,
        "load_seconds": 16.205,
        "size_gb": 1.7039,
        "vram_gb": 1.7039,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite4-3b",
      "display_name": "Granite4 3B",
      "model_version_id": 125,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 74.1381,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 71.2668,
        "decode_tps_c4": 177.132,
        "ttft_s": 0.0602,
        "prefill_tps_4k": 1586.1278,
        "prefill_tps_12k": 1229.4174,
        "load_seconds": 48.938,
        "size_gb": 13.1857,
        "vram_gb": 13.1857,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen2-5-coder-0-5b-instruct",
      "display_name": "Qwen2.5 Coder 0.5B Instruct",
      "model_version_id": 161,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 0.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 32.8247,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 96.6871,
        "decode_tps_c4": 172.544,
        "ttft_s": 0.0453,
        "prefill_tps_4k": 3051.7329,
        "prefill_tps_12k": 2335.6154,
        "load_seconds": 13.6408,
        "size_gb": 2.2186,
        "vram_gb": 2.2186,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "nemotron-mini-4b-instruct",
      "display_name": "Nemotron Mini 4B Instruct",
      "model_version_id": 142,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 4.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 51.9667,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 66.8076,
        "decode_tps_c4": 167.5941,
        "ttft_s": 0.0601,
        "prefill_tps_4k": 2233.8113,
        "prefill_tps_12k": 2567.3169,
        "load_seconds": 21.109,
        "size_gb": 6.6401,
        "vram_gb": 6.6401,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "cogito-3b-v1-preview-llama",
      "display_name": "Cogito 3B V1 Preview Llama",
      "model_version_id": 79,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": "Llama 3.2",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 70.3849,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 73.3802,
        "decode_tps_c4": 157.9272,
        "ttft_s": 0.0539,
        "prefill_tps_4k": 2152.74,
        "prefill_tps_12k": 1600.5926,
        "load_seconds": 68.829,
        "size_gb": 17.4131,
        "vram_gb": 17.4131,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "ministral-3-3b-instruct-2512",
      "display_name": "Ministral 3 3B Instruct 2512",
      "model_version_id": 138,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 80.9089,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 70.8715,
        "decode_tps_c4": 156.7907,
        "ttft_s": 0.0608,
        "prefill_tps_4k": 2515.9526,
        "prefill_tps_12k": 1632.0885,
        "load_seconds": 75.094,
        "size_gb": 16.5584,
        "vram_gb": 16.5584,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-1-moe-1b-instruct",
      "display_name": "Granite3.1 Moe 1B Instruct",
      "model_version_id": 112,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 43.6252,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 102.6271,
        "decode_tps_c4": 149.4756,
        "ttft_s": 0.0649,
        "prefill_tps_4k": 2095.2439,
        "prefill_tps_12k": 1623.637,
        "load_seconds": 44.8459,
        "size_gb": 7.5097,
        "vram_gb": 7.5097,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "phi4-mini-reasoning-3-8b",
      "display_name": "Phi4 Mini Reasoning 3.8B",
      "model_version_id": 151,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.8,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 70.9406,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 48.8338,
        "decode_tps_c4": 139.1304,
        "ttft_s": 0.0665,
        "prefill_tps_4k": 1945.7809,
        "prefill_tps_12k": 1410.7968,
        "load_seconds": 78.016,
        "size_gb": 20.6932,
        "vram_gb": 20.6932,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen2-0-5b-instruct",
      "display_name": "Qwen2 0.5B Instruct",
      "model_version_id": 157,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 0.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 36.0142,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 93.7428,
        "decode_tps_c4": 137.5347,
        "ttft_s": 0.0505,
        "prefill_tps_4k": 3110.1724,
        "prefill_tps_12k": 2336.2753,
        "load_seconds": 14.0775,
        "size_gb": 2.2186,
        "vram_gb": 2.2186,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite4-350m-h",
      "display_name": "Granite4 350M H",
      "model_version_id": 124,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 0.35,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q8_0",
      "overall": 45.5868,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 55.7803,
        "decode_tps_c4": 128.2196,
        "ttft_s": 0.142,
        "prefill_tps_4k": 1739.4345,
        "prefill_tps_12k": 1756.292,
        "load_seconds": 8.367,
        "size_gb": 1.2489,
        "vram_gb": 1.2489,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "lfm2-5-thinking-1-2b",
      "display_name": "Lfm2.5 Thinking 1.2B",
      "model_version_id": 132,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.2,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 74.3049,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 204.953,
        "decode_tps_c4": 122.0261,
        "ttft_s": 0.0296,
        "prefill_tps_4k": 6054.8439,
        "prefill_tps_12k": 6498.2727,
        "load_seconds": 3.0383,
        "size_gb": 1.0978,
        "vram_gb": 1.0978,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "yi-6b-chat-v1-5",
      "display_name": "YI 6B Chat V1.5",
      "model_version_id": 184,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 6.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 58.0319,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 43.3222,
        "decode_tps_c4": 120.2442,
        "ttft_s": 0.1149,
        "prefill_tps_4k": 1256.1834,
        "prefill_tps_12k": 1353.4998,
        "load_seconds": 16.859,
        "size_gb": 5.8833,
        "vram_gb": 5.8833,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen2-5-0-5b-instruct",
      "display_name": "Qwen2.5 0.5B Instruct",
      "model_version_id": 159,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 0.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 43.3114,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 99.2639,
        "decode_tps_c4": 116.8258,
        "ttft_s": 0.0475,
        "prefill_tps_4k": 3069.2425,
        "prefill_tps_12k": 2326.8192,
        "load_seconds": 13.6418,
        "size_gb": 2.2186,
        "vram_gb": 2.2186,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "sailor2-1b-chat",
      "display_name": "Sailor2 1B Chat",
      "model_version_id": 171,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.0,
      "size_class": "tiny",
      "derived_from": "Qwen 2.5",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 23.4625,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 57.0506,
        "decode_tps_c4": 116.3405,
        "ttft_s": 0.0855,
        "prefill_tps_4k": 1537.8271,
        "prefill_tps_12k": 1148.714,
        "load_seconds": 23.3432,
        "size_gb": 4.0764,
        "vram_gb": 4.0764,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "llama3-2-1b-instruct",
      "display_name": "Llama3.2 1B Instruct",
      "model_version_id": 134,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 58.0189,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 58.5414,
        "decode_tps_c4": 114.5358,
        "ttft_s": 0.0685,
        "prefill_tps_4k": 1602.3985,
        "prefill_tps_12k": 1381.4463,
        "load_seconds": 34.6337,
        "size_gb": 5.4304,
        "vram_gb": 5.4304,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "falcon3-1b-instruct",
      "display_name": "Falcon3 1B Instruct",
      "model_version_id": 190,
      "run_id": "r20261028-bv1-7ce712",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 44.3165,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 51.4068,
        "decode_tps_c4": 107.3623,
        "ttft_s": 0.0801,
        "prefill_tps_4k": 1350.2017,
        "prefill_tps_12k": 1381.3319,
        "load_seconds": 35.6724,
        "size_gb": 1.057,
        "vram_gb": 1.057,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite-code-3b-instruct",
      "display_name": "Granite Code 3B Instruct",
      "model_version_id": 108,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 37.9728,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 71.8242,
        "decode_tps_c4": 107.2588,
        "ttft_s": 0.0658,
        "prefill_tps_4k": 1511.3038,
        "prefill_tps_12k": 1715.0365,
        "load_seconds": 25.438,
        "size_gb": 7.6715,
        "vram_gb": 7.6715,
        "gpu_fraction": 1.0,
        "context_length": 2048.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-0-5b-chat-v1-5",
      "display_name": "Qwen 0.5B Chat V1.5",
      "model_version_id": 152,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 0.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 16.1421,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 99.844,
        "decode_tps_c4": 98.9439,
        "ttft_s": 0.054,
        "prefill_tps_4k": 1951.2259,
        "prefill_tps_12k": 1103.804,
        "load_seconds": 105.1233,
        "size_gb": 13.3685,
        "vram_gb": 13.3685,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-2-8b-instruct",
      "display_name": "Granite3.2 8B Instruct",
      "model_version_id": 115,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 74.1207,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 34.6716,
        "decode_tps_c4": 97.6726,
        "ttft_s": 0.1334,
        "prefill_tps_4k": 835.7712,
        "prefill_tps_12k": 714.0253,
        "load_seconds": 100.438,
        "size_gb": 26.9319,
        "vram_gb": 26.9319,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-1-dense-8b-instruct",
      "display_name": "Granite3.1 Dense 8B Instruct",
      "model_version_id": 111,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 72.8603,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 34.8098,
        "decode_tps_c4": 97.2367,
        "ttft_s": 0.1288,
        "prefill_tps_4k": 817.7488,
        "prefill_tps_12k": 685.1801,
        "load_seconds": 105.094,
        "size_gb": 26.9807,
        "vram_gb": 26.9807,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "phi3-3-8b-mini-128k-instruct",
      "display_name": "Phi3 3.8B Mini 128K Instruct",
      "model_version_id": 150,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.8,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 62.9269,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 63.9178,
        "decode_tps_c4": 96.0871,
        "ttft_s": 0.0766,
        "prefill_tps_4k": 1122.9805,
        "prefill_tps_12k": 687.4898,
        "load_seconds": 243.844,
        "size_gb": 54.0613,
        "vram_gb": 54.0613,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-1-moe-3b-instruct",
      "display_name": "Granite3.1 Moe 3B Instruct",
      "model_version_id": 113,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 59.7683,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 53.1127,
        "decode_tps_c4": 92.9557,
        "ttft_s": 0.1254,
        "prefill_tps_4k": 1079.3789,
        "prefill_tps_12k": 838.4479,
        "load_seconds": 61.109,
        "size_gb": 10.9422,
        "vram_gb": 10.9422,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "smollm2-360m-instruct",
      "display_name": "Smollm2 360M Instruct",
      "model_version_id": 176,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 0.36,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 28.0758,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 275.3464,
        "decode_tps_c4": 91.1657,
        "ttft_s": 0.0152,
        "prefill_tps_4k": 9527.2994,
        "prefill_tps_12k": 2732.1585,
        "load_seconds": 13.6042,
        "size_gb": 3.0812,
        "vram_gb": 3.0812,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "yi-coder-1-5b-chat",
      "display_name": "YI Coder 1.5B Chat",
      "model_version_id": 186,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 26.5009,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 49.4729,
        "decode_tps_c4": 88.7185,
        "ttft_s": 0.1006,
        "prefill_tps_4k": 1051.3441,
        "prefill_tps_12k": 682.2714,
        "load_seconds": 183.9436,
        "size_gb": 26.9278,
        "vram_gb": 26.9278,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "falcon3-1b-instruct",
      "display_name": "Falcon3 1B Instruct",
      "model_version_id": 97,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": null,
      "ranked": false,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 58.9894,
        "decode_tps_c4": 87.3233,
        "ttft_s": 0.0818,
        "prefill_tps_4k": 1301.2234,
        "prefill_tps_12k": 1327.4951,
        "load_seconds": 36.3066,
        "size_gb": 6.0352,
        "vram_gb": 6.0352,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "internlm2-1-8b-chat-v2-5",
      "display_name": "Internlm2 1.8B Chat V2.5",
      "model_version_id": 188,
      "run_id": "r20261028-bv1-2e24cd",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.8,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 51.8842,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 45.5755,
        "decode_tps_c4": 85.5815,
        "ttft_s": 0.0987,
        "prefill_tps_4k": 1085.6514,
        "prefill_tps_12k": 889.1267,
        "load_seconds": 82.2355,
        "size_gb": 1.1724,
        "vram_gb": 1.1724,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-1-8b-chat-v1-5",
      "display_name": "Qwen 1.8B Chat V1.5",
      "model_version_id": 153,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.8,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 24.6994,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 127.3801,
        "decode_tps_c4": 82.5127,
        "ttft_s": 0.0307,
        "prefill_tps_4k": 3774.0783,
        "prefill_tps_12k": 2310.5373,
        "load_seconds": 115.8952,
        "size_gb": 27.0761,
        "vram_gb": 27.0761,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-moe-1b-instruct",
      "display_name": "Granite3 Moe 1B Instruct",
      "model_version_id": 120,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 35.5124,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 69.9666,
        "decode_tps_c4": 82.3603,
        "ttft_s": 0.0791,
        "prefill_tps_4k": 1882.2574,
        "prefill_tps_12k": 2179.6198,
        "load_seconds": 16.6659,
        "size_gb": 2.5995,
        "vram_gb": 2.3278,
        "gpu_fraction": 0.8955,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "gemma3-1b-it",
      "display_name": "Gemma3 1B IT",
      "model_version_id": 103,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 56.8376,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 56.9996,
        "decode_tps_c4": 80.0814,
        "ttft_s": 0.1024,
        "prefill_tps_4k": 1928.8723,
        "prefill_tps_12k": 1987.208,
        "load_seconds": 11.268,
        "size_gb": 1.5463,
        "vram_gb": 1.5463,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen2-5-coder-1-5b-instruct",
      "display_name": "Qwen2.5 Coder 1.5B Instruct",
      "model_version_id": 162,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 60.5126,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 45.7585,
        "decode_tps_c4": 77.0294,
        "ttft_s": 0.1006,
        "prefill_tps_4k": 1221.423,
        "prefill_tps_12k": 1016.0871,
        "load_seconds": 26.8341,
        "size_gb": 5.0759,
        "vram_gb": 5.0759,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "internlm2-1-8b-chat-v2-5",
      "display_name": "Internlm2 1.8B Chat V2.5",
      "model_version_id": 129,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.8,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": null,
      "ranked": false,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 43.041,
        "decode_tps_c4": 73.2188,
        "ttft_s": 0.1045,
        "prefill_tps_4k": 1048.7912,
        "prefill_tps_12k": 822.4664,
        "load_seconds": 81.9965,
        "size_gb": 14.2841,
        "vram_gb": 14.2841,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-4b-chat-v1-5",
      "display_name": "Qwen 4B Chat V1.5",
      "model_version_id": 155,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 4.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 30.1605,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 67.4728,
        "decode_tps_c4": 72.1522,
        "ttft_s": 0.0675,
        "prefill_tps_4k": 1750.3491,
        "prefill_tps_12k": 1243.2708,
        "load_seconds": 249.203,
        "size_gb": 56.0779,
        "vram_gb": 56.0779,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepcoder-1-5b-preview",
      "display_name": "Deepcoder 1.5B Preview",
      "model_version_id": 82,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.5,
      "size_class": "tiny",
      "derived_from": "DeepSeek-R1-Distill-Qwen 1.5B",
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 55.8294,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 44.8693,
        "decode_tps_c4": 66.8844,
        "ttft_s": 0.1035,
        "prefill_tps_4k": 1294.875,
        "prefill_tps_12k": 1006.5976,
        "load_seconds": 26.7731,
        "size_gb": 5.0759,
        "vram_gb": 5.0759,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "exaone-deep-2-4b",
      "display_name": "Exaone Deep 2.4B",
      "model_version_id": 92,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.4,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 54.6755,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 31.1631,
        "decode_tps_c4": 65.082,
        "ttft_s": 0.1438,
        "prefill_tps_4k": 776.836,
        "prefill_tps_12k": 615.3835,
        "load_seconds": 76.516,
        "size_gb": 11.8868,
        "vram_gb": 11.8868,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-coder-1-3b-instruct",
      "display_name": "Deepseek Coder 1.3B Instruct",
      "model_version_id": 85,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.3,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 22.6117,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 53.6227,
        "decode_tps_c4": 64.1043,
        "ttft_s": 0.1016,
        "prefill_tps_4k": 1056.5058,
        "prefill_tps_12k": 687.7793,
        "load_seconds": 182.5082,
        "size_gb": 26.8745,
        "vram_gb": 26.8745,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "stablelm2-1-6b-chat",
      "display_name": "Stablelm2 1.6B Chat",
      "model_version_id": 180,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.6,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 24.3833,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 47.8704,
        "decode_tps_c4": 62.2439,
        "ttft_s": 0.0986,
        "prefill_tps_4k": 793.7922,
        "prefill_tps_12k": 981.995,
        "load_seconds": 44.6845,
        "size_gb": 7.4659,
        "vram_gb": 7.4659,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen2-math-1-5b-instruct",
      "display_name": "Qwen2 Math 1.5B Instruct",
      "model_version_id": 193,
      "run_id": "r20261028-bv1-ef0895",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": "math",
      "quantization": "Q4_K_M",
      "overall": 33.1358,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 45.2311,
        "decode_tps_c4": 62.0286,
        "ttft_s": 0.103,
        "prefill_tps_4k": 1216.6652,
        "prefill_tps_12k": 1324.363,
        "load_seconds": 11.5905,
        "size_gb": 0.986,
        "vram_gb": 0.986,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen2-math-1-5b-instruct",
      "display_name": "Qwen2 Math 1.5B Instruct",
      "model_version_id": 163,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": "math",
      "quantization": "Q4_K_M",
      "overall": null,
      "ranked": false,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 44.1416,
        "decode_tps_c4": 61.0823,
        "ttft_s": 0.1041,
        "prefill_tps_4k": 1220.2629,
        "prefill_tps_12k": 1310.6839,
        "load_seconds": 9.1967,
        "size_gb": 2.0633,
        "vram_gb": 2.0633,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "falcon3-3b-instruct",
      "display_name": "Falcon3 3B Instruct",
      "model_version_id": 191,
      "run_id": "r20261028-bv1-7ce712",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 64.8494,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 27.9215,
        "decode_tps_c4": 60.3082,
        "ttft_s": 0.1543,
        "prefill_tps_4k": 673.5075,
        "prefill_tps_12k": 517.9889,
        "load_seconds": 82.8395,
        "size_gb": 2.0057,
        "vram_gb": 2.0057,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-5-14b-instruct",
      "display_name": "Qwen2.5-14B-Instruct",
      "model_version_id": 32,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 14.77,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 91.2781,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 21.67,
        "decode_tps_c4": 59.0027,
        "ttft_s": 0.1831,
        "prefill_tps_4k": 555.2244,
        "prefill_tps_12k": 415.7516,
        "load_seconds": 141.0471,
        "size_gb": 34.8467,
        "vram_gb": 34.8467,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "lfm3-vl-2b",
      "display_name": "Lfm3 VL 2B",
      "model_version_id": 216,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 72.2366,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 41.8658,
        "decode_tps_c4": 58.6535,
        "ttft_s": 0.1503,
        "prefill_tps_4k": 1285.0103,
        "prefill_tps_12k": 1259.0864,
        "load_seconds": 17.3012,
        "size_gb": 2.0083,
        "vram_gb": 2.0083,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-3-2b",
      "display_name": "Granite3.3 2B",
      "model_version_id": 116,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 68.642,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 28.5526,
        "decode_tps_c4": 58.5143,
        "ttft_s": 0.1864,
        "prefill_tps_4k": 645.0414,
        "prefill_tps_12k": 546.4601,
        "load_seconds": 77.968,
        "size_gb": 12.6166,
        "vram_gb": 12.6166,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "lfm3-1-7b",
      "display_name": "Lfm3 1.7B",
      "model_version_id": 215,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.7,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 70.2464,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 42.1177,
        "decode_tps_c4": 58.1799,
        "ttft_s": 0.1509,
        "prefill_tps_4k": 1275.5013,
        "prefill_tps_12k": 1267.8843,
        "load_seconds": 18.957,
        "size_gb": 1.6181,
        "vram_gb": 1.6181,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "falcon3-3b-instruct",
      "display_name": "Falcon3 3B Instruct",
      "model_version_id": 98,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": null,
      "ranked": false,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 25.9809,
        "decode_tps_c4": 57.8924,
        "ttft_s": 0.154,
        "prefill_tps_4k": 685.4602,
        "prefill_tps_12k": 522.43,
        "load_seconds": 93.828,
        "size_gb": 13.9799,
        "vram_gb": 13.9799,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-1-dense-2b-instruct",
      "display_name": "Granite3.1 Dense 2B Instruct",
      "model_version_id": 110,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 60.5775,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 28.7972,
        "decode_tps_c4": 56.7879,
        "ttft_s": 0.1845,
        "prefill_tps_4k": 634.8105,
        "prefill_tps_12k": 534.1775,
        "load_seconds": 70.031,
        "size_gb": 12.641,
        "vram_gb": 12.641,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "smollm2-1-7b-instruct",
      "display_name": "Smollm2 1.7B Instruct",
      "model_version_id": 174,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.7,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 47.0989,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 41.7257,
        "decode_tps_c4": 53.2561,
        "ttft_s": 0.1093,
        "prefill_tps_4k": 670.9104,
        "prefill_tps_12k": 678.9333,
        "load_seconds": 95.0693,
        "size_gb": 14.2408,
        "vram_gb": 14.2408,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "smallthinker-3b-preview",
      "display_name": "Smallthinker 3B Preview",
      "model_version_id": 173,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": "Qwen 2.5 3B",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 61.1071,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 22.113,
        "decode_tps_c4": 53.1948,
        "ttft_s": 0.2076,
        "prefill_tps_4k": 672.4052,
        "prefill_tps_12k": 587.2243,
        "load_seconds": 46.672,
        "size_gb": 7.1606,
        "vram_gb": 7.1606,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "codegemma-7b-instruct-v1-1",
      "display_name": "Codegemma 7B Instruct V1.1",
      "model_version_id": 74,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 59.5044,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.2962,
        "decode_tps_c4": 52.5964,
        "ttft_s": 0.4963,
        "prefill_tps_4k": 198.9778,
        "prefill_tps_12k": 355.9046,
        "load_seconds": 256.719,
        "size_gb": 36.1267,
        "vram_gb": 36.1267,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-5-vl-3b-instruct",
      "display_name": "Qwen2.5-VL-3B-Instruct",
      "model_version_id": 41,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.75,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 72.3435,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 21.9232,
        "decode_tps_c4": 51.4279,
        "ttft_s": 0.2082,
        "prefill_tps_4k": 658.9922,
        "prefill_tps_12k": 574.4994,
        "load_seconds": 24.6148,
        "size_gb": 5.4668,
        "vram_gb": 5.4668,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "huggingfacetb-smollm3-3b",
      "display_name": "SmolLM3-3B",
      "model_version_id": 14,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.08,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 78.0448,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 22.2367,
        "decode_tps_c4": 50.7828,
        "ttft_s": 0.1972,
        "prefill_tps_4k": 684.0474,
        "prefill_tps_12k": 594.8398,
        "load_seconds": 75.7615,
        "size_gb": 11.9592,
        "vram_gb": 11.9592,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "hermes3-3b-llama3-2",
      "display_name": "Hermes3 3B Llama3.2",
      "model_version_id": 128,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": "Llama 3.2",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 48.4727,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 23.3722,
        "decode_tps_c4": 50.7232,
        "ttft_s": 0.193,
        "prefill_tps_4k": 501.8384,
        "prefill_tps_12k": 375.8361,
        "load_seconds": 106.875,
        "size_gb": 17.4131,
        "vram_gb": 17.4131,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-r1-1-5b-qwen-distill",
      "display_name": "Deepseek R1 1.5B Qwen Distill",
      "model_version_id": 88,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 48.918,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 44.8517,
        "decode_tps_c4": 50.5694,
        "ttft_s": 0.1019,
        "prefill_tps_4k": 1286.6169,
        "prefill_tps_12k": 1006.2381,
        "load_seconds": 26.9672,
        "size_gb": 5.0759,
        "vram_gb": 5.0759,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite4-2-3b",
      "display_name": "Granite4.2 3B",
      "model_version_id": 209,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 78.2385,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 24.2936,
        "decode_tps_c4": 50.144,
        "ttft_s": 0.1966,
        "prefill_tps_4k": 472.0694,
        "prefill_tps_12k": 346.6334,
        "load_seconds": 84.5551,
        "size_gb": 13.1857,
        "vram_gb": 13.1857,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite4-1b-h",
      "display_name": "Granite4 1B H",
      "model_version_id": 123,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q8_0",
      "overall": 66.7905,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 24.9159,
        "decode_tps_c4": 50.142,
        "ttft_s": 0.3467,
        "prefill_tps_4k": 615.6944,
        "prefill_tps_12k": 631.2839,
        "load_seconds": 15.75,
        "size_gb": 3.3565,
        "vram_gb": 3.3565,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen2-1-5b-instruct",
      "display_name": "Qwen2 1.5B Instruct",
      "model_version_id": 158,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 49.6912,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 43.0686,
        "decode_tps_c4": 49.7986,
        "ttft_s": 0.118,
        "prefill_tps_4k": 1284.0797,
        "prefill_tps_12k": 1000.4801,
        "load_seconds": null,
        "size_gb": null,
        "vram_gb": null,
        "gpu_fraction": null,
        "context_length": null
      },
      "perf_units": {
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "ttft_s": "s"
      }
    },
    {
      "slug": "meta-llama-llama-3-2-3b-instruct",
      "display_name": "Llama-3.2-3B-Instruct",
      "model_version_id": 19,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.21,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 75.7368,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 20.6795,
        "decode_tps_c4": 49.7549,
        "ttft_s": 0.2033,
        "prefill_tps_4k": 497.209,
        "prefill_tps_12k": 379.5428,
        "load_seconds": 67.8224,
        "size_gb": 9.8717,
        "vram_gb": 9.8717,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-5-coder-3b-instruct",
      "display_name": "Qwen2.5-Coder-3B-Instruct",
      "model_version_id": 38,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.09,
      "size_class": "le4",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 72.5305,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 21.3091,
        "decode_tps_c4": 49.332,
        "ttft_s": 0.2076,
        "prefill_tps_4k": 665.1641,
        "prefill_tps_12k": 579.0787,
        "load_seconds": 22.7514,
        "size_gb": 4.7279,
        "vram_gb": 4.7279,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "ibm-granite-granite-4-1-3b",
      "display_name": "granite-4.1-3b",
      "model_version_id": 15,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.4,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 80.7146,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 20.1362,
        "decode_tps_c4": 49.0449,
        "ttft_s": 0.246,
        "prefill_tps_4k": 472.0635,
        "prefill_tps_12k": 345.8545,
        "load_seconds": 48.167,
        "size_gb": 7.796,
        "vram_gb": 7.796,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-5-3b-instruct",
      "display_name": "Qwen2.5-3B-Instruct",
      "model_version_id": 34,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.09,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 76.7703,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 8,
      "concurrency_text": "4 to 8 at a time",
      "perf": {
        "decode_tps": 21.8785,
        "decode_tps_c4": 44.9175,
        "ttft_s": 0.2039,
        "prefill_tps_4k": 680.6535,
        "prefill_tps_12k": 591.14,
        "load_seconds": 23.3712,
        "size_gb": 4.7279,
        "vram_gb": 4.7279,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "microsoft-phi-3-mini-4k-instruct",
      "display_name": "Phi-3-mini-4k-instruct",
      "model_version_id": 22,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.82,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 63.4316,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 17.6273,
        "decode_tps_c4": 44.5007,
        "ttft_s": 0.2745,
        "prefill_tps_4k": 329.0263,
        "prefill_tps_12k": 371.1952,
        "load_seconds": 57.7854,
        "size_gb": 8.9138,
        "vram_gb": 8.9138,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen2-5-1-5b-instruct",
      "display_name": "Qwen2.5 1.5B Instruct",
      "model_version_id": 160,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 68.5148,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 45.4934,
        "decode_tps_c4": 43.8323,
        "ttft_s": 0.1,
        "prefill_tps_4k": 1222.1021,
        "prefill_tps_12k": 1003.4702,
        "load_seconds": 27.036,
        "size_gb": 5.0759,
        "vram_gb": 5.0759,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "microsoft-phi-4-mini-instruct",
      "display_name": "Phi-4-mini-instruct",
      "model_version_id": 24,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.84,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 81.5627,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 17.1375,
        "decode_tps_c4": 43.73,
        "ttft_s": 0.2254,
        "prefill_tps_4k": 514.2015,
        "prefill_tps_12k": 372.5218,
        "load_seconds": 73.2867,
        "size_gb": 11.3924,
        "vram_gb": 11.3924,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-dense-2b-instruct",
      "display_name": "Granite3 Dense 2B Instruct",
      "model_version_id": 118,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 58.3099,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 29.6697,
        "decode_tps_c4": 43.4875,
        "ttft_s": 0.1839,
        "prefill_tps_4k": 625.615,
        "prefill_tps_12k": 652.2115,
        "load_seconds": 20.109,
        "size_gb": 4.3705,
        "vram_gb": 4.3705,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen3-1-7b",
      "display_name": "Qwen3 1.7B",
      "model_version_id": 166,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.7,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 76.0225,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 38.5041,
        "decode_tps_c4": 43.4559,
        "ttft_s": 0.1264,
        "prefill_tps_4k": 1056.7413,
        "prefill_tps_12k": 779.2915,
        "load_seconds": 95.4523,
        "size_gb": 16.4959,
        "vram_gb": 16.4959,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "gemma-2b-instruct-v1-1",
      "display_name": "Gemma 2B Instruct V1.1",
      "model_version_id": 100,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 41.8888,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 29.2977,
        "decode_tps_c4": 43.4303,
        "ttft_s": 0.1584,
        "prefill_tps_4k": 860.6247,
        "prefill_tps_12k": 858.877,
        "load_seconds": 13.453,
        "size_gb": 3.3356,
        "vram_gb": 3.3356,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "codegemma-2b-code-v1-1",
      "display_name": "Codegemma 2B Code V1.1",
      "model_version_id": 73,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 15.2962,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 29.0816,
        "decode_tps_c4": 43.2579,
        "ttft_s": 0.1418,
        "prefill_tps_4k": 858.209,
        "prefill_tps_12k": 856.1354,
        "load_seconds": 10.782,
        "size_gb": 3.3356,
        "vram_gb": 3.3356,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-4b-instruct-2507",
      "display_name": "Qwen3-4B-Instruct-2507",
      "model_version_id": 49,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 4.02,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 90.2603,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 17.0756,
        "decode_tps_c4": 43.182,
        "ttft_s": 0.2618,
        "prefill_tps_4k": 412.3869,
        "prefill_tps_12k": 308.3372,
        "load_seconds": 85.7297,
        "size_gb": 12.5367,
        "vram_gb": 12.5367,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "google-medgemma-4b-it",
      "display_name": "medgemma-4b-it",
      "model_version_id": 13,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 4.3,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": "medical",
      "quantization": "Q4_K_M",
      "overall": 74.3436,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 17.103,
        "decode_tps_c4": 42.184,
        "ttft_s": 0.3097,
        "prefill_tps_4k": 490.1869,
        "prefill_tps_12k": 491.2058,
        "load_seconds": 26.5966,
        "size_gb": 4.4596,
        "vram_gb": 4.4596,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "lfm2-5-8b-a1b",
      "display_name": "Lfm2.5 8B A1B",
      "model_version_id": 131,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 75.7244,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 43.4376,
        "decode_tps_c4": 41.6108,
        "ttft_s": 0.2304,
        "prefill_tps_4k": 941.7579,
        "prefill_tps_12k": 910.1233,
        "load_seconds": 9.5,
        "size_gb": 5.5042,
        "vram_gb": 5.5042,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "microsoft-phi-3-5-mini-instruct",
      "display_name": "Phi-3.5-mini-instruct",
      "model_version_id": 21,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.82,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 67.0529,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 17.5846,
        "decode_tps_c4": 41.5979,
        "ttft_s": 0.2837,
        "prefill_tps_4k": 276.1487,
        "prefill_tps_12k": 156.0935,
        "load_seconds": 219.1701,
        "size_gb": 28.4257,
        "vram_gb": 28.4257,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-4b",
      "display_name": "Qwen3-4B",
      "model_version_id": 48,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 4.02,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 89.1828,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 16.9503,
        "decode_tps_c4": 41.3392,
        "ttft_s": 0.2688,
        "prefill_tps_4k": 483.6555,
        "prefill_tps_12k": 342.9295,
        "load_seconds": 66.5486,
        "size_gb": 12.6602,
        "vram_gb": 12.6602,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite4-tiny-h",
      "display_name": "Granite4 Tiny H",
      "model_version_id": 127,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 79.1049,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 24.2409,
        "decode_tps_c4": 40.9338,
        "ttft_s": 0.4229,
        "prefill_tps_4k": 442.007,
        "prefill_tps_12k": 476.0533,
        "load_seconds": 19.359,
        "size_gb": 6.0593,
        "vram_gb": 6.0593,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepscaler-1-5b-preview",
      "display_name": "Deepscaler 1.5B Preview",
      "model_version_id": 84,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.5,
      "size_class": "tiny",
      "derived_from": "DeepSeek-R1-Distill-Qwen 1.5B",
      "specialised": "math",
      "quantization": "Q4_K_M",
      "overall": 52.0944,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 45.5302,
        "decode_tps_c4": 40.0417,
        "ttft_s": 0.101,
        "prefill_tps_4k": 1287.3196,
        "prefill_tps_12k": 1006.4777,
        "load_seconds": 29.1182,
        "size_gb": 5.0759,
        "vram_gb": 5.0759,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "laguna-xs-2-1",
      "display_name": "Laguna XS 2.1",
      "model_version_id": 213,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": null,
      "size_class": null,
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 91.1849,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 23.1114,
        "decode_tps_c4": 40.0182,
        "ttft_s": 0.642,
        "prefill_tps_4k": 323.363,
        "prefill_tps_12k": 288.9103,
        "load_seconds": 68.3955,
        "size_gb": 23.1855,
        "vram_gb": 23.1855,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "gemma2-2b-instruct",
      "display_name": "Gemma2 2B Instruct",
      "model_version_id": 102,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 68.1774,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 27.5628,
        "decode_tps_c4": 39.8645,
        "ttft_s": 0.1922,
        "prefill_tps_4k": 718.6028,
        "prefill_tps_12k": 716.1897,
        "load_seconds": 45.859,
        "size_gb": 4.6806,
        "vram_gb": 4.6806,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "stablelm-zephyr-3b",
      "display_name": "Stablelm Zephyr 3B",
      "model_version_id": 179,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.8,
      "size_class": "tiny",
      "derived_from": "StableLM 3B",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 43.7042,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 26.539,
        "decode_tps_c4": 39.5994,
        "ttft_s": 0.2076,
        "prefill_tps_4k": 406.6915,
        "prefill_tps_12k": 523.5139,
        "load_seconds": 72.531,
        "size_gb": 12.5071,
        "vram_gb": 12.5071,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "bartowski-spark-x2-6-4b",
      "display_name": "Spark-X2.6-4B",
      "model_version_id": 198,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 4.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 85.0137,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 19.7465,
        "decode_tps_c4": 38.714,
        "ttft_s": 0.3802,
        "prefill_tps_4k": 468.9994,
        "prefill_tps_12k": 438.4377,
        "load_seconds": 58.1939,
        "size_gb": 5.3148,
        "vram_gb": 5.3148,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "google-gemma-3-4b-it",
      "display_name": "gemma-3-4b-it",
      "model_version_id": 8,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 4.3,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 82.0911,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 16.9063,
        "decode_tps_c4": 38.6304,
        "ttft_s": 0.3071,
        "prefill_tps_4k": 492.6016,
        "prefill_tps_12k": 490.4903,
        "load_seconds": 27.1081,
        "size_gb": 4.4594,
        "vram_gb": 4.4594,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "stable-code-3b-instruct",
      "display_name": "Stable Code 3B Instruct",
      "model_version_id": 178,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.8,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 22.808,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 26.4371,
        "decode_tps_c4": 38.6196,
        "ttft_s": 0.2128,
        "prefill_tps_4k": 403.7822,
        "prefill_tps_12k": 224.2395,
        "load_seconds": 316.109,
        "size_gb": 44.905,
        "vram_gb": 44.905,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "dolphin-phi-2-7b-v2-6",
      "display_name": "Dolphin Phi 2.7B V2.6",
      "model_version_id": 91,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.7,
      "size_class": "tiny",
      "derived_from": "Phi-2",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 44.019,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 25.0211,
        "decode_tps_c4": 38.4384,
        "ttft_s": 0.1742,
        "prefill_tps_4k": 502.8911,
        "prefill_tps_12k": 515.4769,
        "load_seconds": 38.828,
        "size_gb": 7.1705,
        "vram_gb": 7.1705,
        "gpu_fraction": 1.0,
        "context_length": 2048.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "phi-2-7b-chat-v2",
      "display_name": "Phi 2.7B Chat V2",
      "model_version_id": 148,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.7,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 39.8792,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 26.0076,
        "decode_tps_c4": 38.2132,
        "ttft_s": 0.1717,
        "prefill_tps_4k": 507.9095,
        "prefill_tps_12k": 528.6954,
        "load_seconds": 37.266,
        "size_gb": 7.1705,
        "vram_gb": 7.1705,
        "gpu_fraction": 1.0,
        "context_length": 2048.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "swiss-ai-apertus-8b-instruct-2509",
      "display_name": "Apertus-8B-Instruct-2509",
      "model_version_id": 195,
      "run_id": "r20261029-bv1-9ddd5f",
      "backend": "Vulkan (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.05,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 63.6104,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 13.5348,
        "decode_tps_c4": 32.9398,
        "ttft_s": 0.3845,
        "prefill_tps_4k": 187.8229,
        "prefill_tps_12k": 115.3327,
        "load_seconds": 10.6447,
        "size_gb": 22.3054,
        "vram_gb": 22.3054,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen3-5-2b",
      "display_name": "Qwen3.5 2B",
      "model_version_id": 167,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 75.0878,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 33.3131,
        "decode_tps_c4": 32.1578,
        "ttft_s": 0.2042,
        "prefill_tps_4k": 961.3919,
        "prefill_tps_12k": 939.157,
        "load_seconds": 8.297,
        "size_gb": 1.8618,
        "vram_gb": 1.8618,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-2-2b-instruct",
      "display_name": "Granite3.2 2B Instruct",
      "model_version_id": 114,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 63.844,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 30.0715,
        "decode_tps_c4": 32.1347,
        "ttft_s": 0.1835,
        "prefill_tps_4k": 635.3461,
        "prefill_tps_12k": 537.8871,
        "load_seconds": 71.4919,
        "size_gb": 12.6166,
        "vram_gb": 12.6166,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-4b-thinking-2507",
      "display_name": "Qwen3-4B-Thinking-2507",
      "model_version_id": 238,
      "run_id": "r20261105-bv1-2abe78",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 4.02,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 83.5194,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 21.4766,
        "decode_tps_c4": 32.1159,
        "ttft_s": 0.2355,
        "prefill_tps_4k": 511.4064,
        "prefill_tps_12k": 352.3157,
        "load_seconds": 126.4797,
        "size_gb": 2.4973,
        "vram_gb": 2.4973,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "flint-40b-a2b",
      "display_name": "Flint 40B A2B",
      "model_version_id": 205,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 40.0,
      "size_class": "gt35",
      "derived_from": null,
      "specialised": null,
      "quantization": "NVFP4",
      "overall": 88.9758,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 25.8941,
        "decode_tps_c4": 31.8972,
        "ttft_s": 0.6642,
        "prefill_tps_4k": 342.457,
        "prefill_tps_12k": 322.8328,
        "load_seconds": 60.3351,
        "size_gb": 24.2801,
        "vram_gb": 24.2801,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "openai-gpt-oss-20b",
      "display_name": "gpt-oss-20b",
      "model_version_id": 237,
      "run_id": "r20261105-bv1-2abe78",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 20.91,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "MXFP4",
      "overall": 95.026,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 14.8812,
        "decode_tps_c4": 31.4864,
        "ttft_s": 0.6209,
        "prefill_tps_4k": 437.1095,
        "prefill_tps_12k": 429.0446,
        "load_seconds": 40.5584,
        "size_gb": 13.5285,
        "vram_gb": 13.5285,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite4-micro-h",
      "display_name": "Granite4 Micro H",
      "model_version_id": 126,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 3.0,
      "size_class": "le4",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 76.6382,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 13.3735,
        "decode_tps_c4": 30.9272,
        "ttft_s": 0.6006,
        "prefill_tps_4k": 364.4477,
        "prefill_tps_12k": 369.9609,
        "load_seconds": 23.938,
        "size_gb": 3.9841,
        "vram_gb": 3.9841,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "medgemma1-5-4b-it",
      "display_name": "Medgemma1.5 4B IT",
      "model_version_id": 137,
      "run_id": "r20261029-bv1-3e4075",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 4.0,
      "size_class": "le4",
      "derived_from": "Gemma 3",
      "specialised": "medical",
      "quantization": "Q4_K_M",
      "overall": 71.5704,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 20.1878,
        "decode_tps_c4": 30.6431,
        "ttft_s": 0.2759,
        "prefill_tps_4k": 508.4888,
        "prefill_tps_12k": 507.9796,
        "load_seconds": 39.3504,
        "size_gb": 5.9444,
        "vram_gb": 5.9444,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "openthinker-7b-v2",
      "display_name": "Openthinker 7B V2",
      "model_version_id": 146,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": "Qwen 2.5",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 76.7682,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 12.5576,
        "decode_tps_c4": 30.0761,
        "ttft_s": 0.3586,
        "prefill_tps_4k": 307.5555,
        "prefill_tps_12k": 280.2105,
        "load_seconds": 59.593,
        "size_gb": 12.5369,
        "vram_gb": 12.5369,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "gemma4-e2b-it",
      "display_name": "Gemma4 E2B IT",
      "model_version_id": 105,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 5.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 93.7501,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 17.153,
        "decode_tps_c4": 29.816,
        "ttft_s": 0.3947,
        "prefill_tps_4k": 672.8474,
        "prefill_tps_12k": 593.0239,
        "load_seconds": 18.344,
        "size_gb": 0.5929,
        "vram_gb": 0.5929,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "lfm2-24b",
      "display_name": "Lfm2 24B",
      "model_version_id": 214,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 24.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 87.3294,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 32.4672,
        "decode_tps_c4": 29.4636,
        "ttft_s": 0.5077,
        "prefill_tps_4k": 583.0121,
        "prefill_tps_12k": 591.1745,
        "load_seconds": 16.1083,
        "size_gb": 14.9609,
        "vram_gb": 14.9609,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mistralai-mistral-7b-instruct-v0-2",
      "display_name": "Mistral-7B-Instruct-v0.2",
      "model_version_id": 27,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.24,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 59.2136,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.2045,
        "decode_tps_c4": 29.0183,
        "ttft_s": 0.4304,
        "prefill_tps_4k": 274.1602,
        "prefill_tps_12k": 224.9593,
        "load_seconds": 131.8757,
        "size_gb": 21.994,
        "vram_gb": 21.994,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "exaone-deep-7-8b",
      "display_name": "Exaone Deep 7.8B",
      "model_version_id": 93,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.8,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 65.7065,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 12.1798,
        "decode_tps_c4": 28.8573,
        "ttft_s": 0.3663,
        "prefill_tps_4k": 275.5567,
        "prefill_tps_12k": 229.5328,
        "load_seconds": 131.422,
        "size_gb": 22.2641,
        "vram_gb": 22.2641,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "falcon3-7b-instruct",
      "display_name": "Falcon3 7B Instruct",
      "model_version_id": 192,
      "run_id": "r20261028-bv1-7ce712",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 76.2343,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 12.507,
        "decode_tps_c4": 28.4187,
        "ttft_s": 0.3699,
        "prefill_tps_4k": 290.0792,
        "prefill_tps_12k": 249.6633,
        "load_seconds": 103.4836,
        "size_gb": 4.5697,
        "vram_gb": 4.5697,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-5-vl-7b-instruct",
      "display_name": "Qwen2.5-VL-7B-Instruct",
      "model_version_id": 42,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.29,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 82.0256,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.2455,
        "decode_tps_c4": 28.35,
        "ttft_s": 0.3966,
        "prefill_tps_4k": 305.576,
        "prefill_tps_12k": 277.6815,
        "load_seconds": 63.9528,
        "size_gb": 13.2466,
        "vram_gb": 13.2466,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-5-7b-instruct",
      "display_name": "Qwen2.5-7B-Instruct",
      "model_version_id": 35,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.62,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 85.5675,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 10.1198,
        "decode_tps_c4": 28.2939,
        "ttft_s": 0.3928,
        "prefill_tps_4k": 306.2546,
        "prefill_tps_12k": 278.8864,
        "load_seconds": 34.1411,
        "size_gb": 8.7495,
        "vram_gb": 8.7495,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "internlm2-7b-chat-v2-5",
      "display_name": "Internlm2 7B Chat V2.5",
      "model_version_id": 130,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 67.3585,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 12.0246,
        "decode_tps_c4": 28.2607,
        "ttft_s": 0.4145,
        "prefill_tps_4k": 252.2024,
        "prefill_tps_12k": 217.1607,
        "load_seconds": 135.297,
        "size_gb": 22.231,
        "vram_gb": 22.231,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "tulu3-8b",
      "display_name": "Tulu3 8B",
      "model_version_id": 183,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": "Llama 3.1",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 72.9871,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 12.0478,
        "decode_tps_c4": 28.2358,
        "ttft_s": 0.3732,
        "prefill_tps_4k": 273.9014,
        "prefill_tps_12k": 231.1394,
        "load_seconds": 126.469,
        "size_gb": 22.351,
        "vram_gb": 22.351,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-5-coder-7b-instruct",
      "display_name": "Qwen2.5-Coder-7B-Instruct",
      "model_version_id": 39,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.62,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 84.4797,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 9.8024,
        "decode_tps_c4": 28.2216,
        "ttft_s": 0.396,
        "prefill_tps_4k": 307.7871,
        "prefill_tps_12k": 279.3404,
        "load_seconds": 62.7367,
        "size_gb": 12.5076,
        "vram_gb": 12.5076,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "meta-llama-meta-llama-3-8b-instruct",
      "display_name": "Meta-Llama-3-8B-Instruct",
      "model_version_id": 20,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.03,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 74.7278,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 9.5772,
        "decode_tps_c4": 27.2717,
        "ttft_s": 0.4067,
        "prefill_tps_4k": 272.2075,
        "prefill_tps_12k": 271.2225,
        "load_seconds": 38.1772,
        "size_gb": 9.1557,
        "vram_gb": 9.1557,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "codellama-7b-instruct",
      "display_name": "Codellama 7B Instruct",
      "model_version_id": 76,
      "run_id": "r20261022-bv1-54a19b",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 53.0242,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 13.3667,
        "decode_tps_c4": 27.182,
        "ttft_s": 0.4032,
        "prefill_tps_4k": 252.3193,
        "prefill_tps_12k": 181.3559,
        "load_seconds": 248.406,
        "size_gb": 38.8382,
        "vram_gb": 38.8382,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "meta-llama-llama-2-7b-chat-hf",
      "display_name": "Llama-2-7b-chat-hf",
      "model_version_id": 17,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 6.74,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 43.7532,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 8,
      "concurrency_text": "4 to 8 at a time",
      "perf": {
        "decode_tps": 10.9642,
        "decode_tps_c4": 26.7272,
        "ttft_s": 0.4311,
        "prefill_tps_4k": 266.1857,
        "prefill_tps_12k": 291.4854,
        "load_seconds": 68.3621,
        "size_gb": 12.7905,
        "vram_gb": 12.7905,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen3-vl-2b-thinking",
      "display_name": "Qwen3 VL 2B Thinking",
      "model_version_id": 169,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 53.7037,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 27.151,
        "decode_tps_c4": 26.4367,
        "ttft_s": 0.1442,
        "prefill_tps_4k": 949.6726,
        "prefill_tps_12k": 806.4299,
        "load_seconds": 19.016,
        "size_gb": 3.5949,
        "vram_gb": 3.5949,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "sailor2-8b-chat",
      "display_name": "Sailor2 8B Chat",
      "model_version_id": 172,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": "Qwen 2.5",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 57.2899,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.5903,
        "decode_tps_c4": 26.2881,
        "ttft_s": 0.4249,
        "prefill_tps_4k": 223.2833,
        "prefill_tps_12k": 204.3435,
        "load_seconds": 86.312,
        "size_gb": 14.1708,
        "vram_gb": 14.1708,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen3-vl-2b-instruct",
      "display_name": "Qwen3 VL 2B Instruct",
      "model_version_id": 168,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 2.0,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 77.364,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 26.4991,
        "decode_tps_c4": 26.2147,
        "ttft_s": 0.145,
        "prefill_tps_4k": 933.3264,
        "prefill_tps_12k": 787.9321,
        "load_seconds": 16.532,
        "size_gb": 3.5949,
        "vram_gb": 3.5949,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-30b-a3b-instruct-2507",
      "display_name": "Qwen3-30B-A3B-Instruct-2507",
      "model_version_id": 46,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 30.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 95.0349,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 20.36,
        "decode_tps_c4": 26.1234,
        "ttft_s": 0.4904,
        "prefill_tps_4k": 383.5951,
        "prefill_tps_12k": 303.9226,
        "load_seconds": 100.2712,
        "size_gb": 31.6801,
        "vram_gb": 31.6801,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "gemma3n-e2b-it",
      "display_name": "Gemma3n E2B IT",
      "model_version_id": 104,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 5.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 79.8536,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 13.816,
        "decode_tps_c4": 26.0991,
        "ttft_s": 0.3533,
        "prefill_tps_4k": 483.8795,
        "prefill_tps_12k": 486.1134,
        "load_seconds": 19.593,
        "size_gb": 3.3972,
        "vram_gb": 3.3972,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-coder-30b-a3b-instruct",
      "display_name": "Qwen3-Coder-30B-A3B-Instruct",
      "model_version_id": 58,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 30.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 95.5399,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 20.0553,
        "decode_tps_c4": 26.0777,
        "ttft_s": 0.4889,
        "prefill_tps_4k": 382.3662,
        "prefill_tps_12k": 302.973,
        "load_seconds": 100.4754,
        "size_gb": 31.6801,
        "vram_gb": 31.6801,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-3-8b",
      "display_name": "Granite3.3 8B",
      "model_version_id": 117,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 79.395,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 11.3445,
        "decode_tps_c4": 25.8932,
        "ttft_s": 0.4756,
        "prefill_tps_4k": 254.4203,
        "prefill_tps_12k": 203.4872,
        "load_seconds": 154.485,
        "size_gb": 26.9319,
        "vram_gb": 26.9319,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-30b-a3b",
      "display_name": "Qwen3-30B-A3B",
      "model_version_id": 45,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 30.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.7502,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 19.6786,
        "decode_tps_c4": 25.4957,
        "ttft_s": 0.5183,
        "prefill_tps_4k": 379.7907,
        "prefill_tps_12k": 301.0756,
        "load_seconds": 101.3695,
        "size_gb": 31.7459,
        "vram_gb": 31.7459,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "gemma4-e4b-it",
      "display_name": "Gemma4 E4B IT",
      "model_version_id": 106,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 95.4613,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 13.1584,
        "decode_tps_c4": 25.3133,
        "ttft_s": 0.4079,
        "prefill_tps_4k": 416.4618,
        "prefill_tps_12k": 390.672,
        "load_seconds": 30.469,
        "size_gb": 1.1893,
        "vram_gb": 1.1893,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "codeqwen-7b-chat-v1-5",
      "display_name": "Codeqwen 7B Chat V1.5",
      "model_version_id": 77,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 49.3922,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 11.8161,
        "decode_tps_c4": 25.2161,
        "ttft_s": 0.3825,
        "prefill_tps_4k": 315.6289,
        "prefill_tps_12k": 278.861,
        "load_seconds": 68.796,
        "size_gb": 13.6452,
        "vram_gb": 13.6452,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite4-2-8b",
      "display_name": "Granite4.2 8B",
      "model_version_id": 210,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 81.8653,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 11.062,
        "decode_tps_c4": 24.8466,
        "ttft_s": 0.4154,
        "prefill_tps_4k": 243.6117,
        "prefill_tps_12k": 196.7583,
        "load_seconds": 169.1508,
        "size_gb": 27.1039,
        "vram_gb": 27.1039,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-llm-7b-chat",
      "display_name": "Deepseek Llm 7B Chat",
      "model_version_id": 87,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 54.3309,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 12.2261,
        "decode_tps_c4": 24.8056,
        "ttft_s": 0.3661,
        "prefill_tps_4k": 278.355,
        "prefill_tps_12k": 299.0783,
        "load_seconds": 118.078,
        "size_gb": 20.3004,
        "vram_gb": 20.3004,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "falcon3-7b-instruct",
      "display_name": "Falcon3 7B Instruct",
      "model_version_id": 99,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": null,
      "ranked": false,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 11.6426,
        "decode_tps_c4": 24.7027,
        "ttft_s": 0.3756,
        "prefill_tps_4k": 278.6717,
        "prefill_tps_12k": 240.2589,
        "load_seconds": 112.218,
        "size_gb": 20.1049,
        "vram_gb": 20.1049,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "cogito-8b-v1-preview-llama",
      "display_name": "Cogito 8B V1 Preview Llama",
      "model_version_id": 80,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": "Llama 3.1",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 83.8743,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 11.2234,
        "decode_tps_c4": 24.6189,
        "ttft_s": 0.3727,
        "prefill_tps_4k": 275.4079,
        "prefill_tps_12k": 226.1676,
        "load_seconds": 128.875,
        "size_gb": 22.351,
        "vram_gb": 22.351,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "yi-coder-9b-chat",
      "display_name": "YI Coder 9B Chat",
      "model_version_id": 187,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 9.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 59.0417,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.4522,
        "decode_tps_c4": 24.6006,
        "ttft_s": 0.5387,
        "prefill_tps_4k": 218.812,
        "prefill_tps_12k": 188.8952,
        "load_seconds": 98.703,
        "size_gb": 18.5368,
        "vram_gb": 18.5368,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "yi-9b-chat-v1-5",
      "display_name": "YI 9B Chat V1.5",
      "model_version_id": 185,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 9.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 70.474,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 9.9561,
        "decode_tps_c4": 24.5729,
        "ttft_s": 0.5477,
        "prefill_tps_4k": 229.0683,
        "prefill_tps_12k": 242.0483,
        "load_seconds": 33.0,
        "size_gb": 8.612,
        "vram_gb": 8.612,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "allenai-olmo-3-7b-think",
      "display_name": "Olmo-3-7B-Think",
      "model_version_id": 229,
      "run_id": "r20261105-bv1-2abe78",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.3,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 77.7496,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 13.4627,
        "decode_tps_c4": 23.9585,
        "ttft_s": 0.6121,
        "prefill_tps_4k": 256.7173,
        "prefill_tps_12k": 229.1374,
        "load_seconds": 209.8619,
        "size_gb": 4.472,
        "vram_gb": 4.472,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "marco-o1-7b",
      "display_name": "Marco O1 7B",
      "model_version_id": 135,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": "Qwen 2 7B",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 70.4729,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 11.8004,
        "decode_tps_c4": 23.7108,
        "ttft_s": 0.3707,
        "prefill_tps_4k": 315.7247,
        "prefill_tps_12k": 275.6601,
        "load_seconds": 58.39,
        "size_gb": 12.5369,
        "vram_gb": 12.5369,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "meta-llama-llama-3-1-8b-instruct",
      "display_name": "Llama-3.1-8B-Instruct",
      "model_version_id": 18,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.03,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 81.017,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 9.551,
        "decode_tps_c4": 23.5598,
        "ttft_s": 0.406,
        "prefill_tps_4k": 278.2606,
        "prefill_tps_12k": 228.8605,
        "load_seconds": 70.082,
        "size_gb": 13.7275,
        "vram_gb": 13.7275,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-7b-chat-v1-5",
      "display_name": "Qwen 7B Chat V1.5",
      "model_version_id": 156,
      "run_id": "r20261022-bv1-54a19b",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 44.091,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 12.3461,
        "decode_tps_c4": 23.1981,
        "ttft_s": 0.4216,
        "prefill_tps_4k": 249.4216,
        "prefill_tps_12k": 183.8565,
        "load_seconds": 256.75,
        "size_gb": 39.2427,
        "vram_gb": 39.2427,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "lgai-exaone-exaone-3-5-7-8b-instruct",
      "display_name": "EXAONE-3.5-7.8B-Instruct",
      "model_version_id": 16,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.82,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 82.0459,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 9.4804,
        "decode_tps_c4": 23.0027,
        "ttft_s": 0.3985,
        "prefill_tps_4k": 277.5624,
        "prefill_tps_12k": 228.5496,
        "load_seconds": 70.1358,
        "size_gb": 13.6406,
        "vram_gb": 13.6406,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "allenai-olmo-3-7b-instruct",
      "display_name": "Olmo-3-7B-Instruct",
      "model_version_id": 1,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.3,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 82.8567,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 8,
      "concurrency_text": "4 to 8 at a time",
      "perf": {
        "decode_tps": 10.2493,
        "decode_tps_c4": 22.758,
        "ttft_s": 0.6463,
        "prefill_tps_4k": 234.673,
        "prefill_tps_12k": 194.6145,
        "load_seconds": 141.3684,
        "size_gb": 14.3642,
        "vram_gb": 14.3642,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-coder-6-7b-instruct",
      "display_name": "Deepseek Coder 6.7B Instruct",
      "model_version_id": 86,
      "run_id": "r20261022-bv1-54a19b",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 6.7,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 34.6862,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 13.5797,
        "decode_tps_c4": 22.5,
        "ttft_s": 0.4079,
        "prefill_tps_4k": 277.2862,
        "prefill_tps_12k": 196.6318,
        "load_seconds": 262.032,
        "size_gb": 38.8391,
        "vram_gb": 38.8391,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "ministral-3-8b-instruct-2512",
      "display_name": "Ministral 3 8B Instruct 2512",
      "model_version_id": 139,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 88.7303,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.9993,
        "decode_tps_c4": 22.1928,
        "ttft_s": 0.4141,
        "prefill_tps_4k": 270.9169,
        "prefill_tps_12k": 200.5405,
        "load_seconds": 145.25,
        "size_gb": 23.7624,
        "vram_gb": 23.7624,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "google-gemma-4-26b-a4b-it",
      "display_name": "gemma-4-26B-A4B-it",
      "model_version_id": 11,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 26.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 97.7523,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 15.7103,
        "decode_tps_c4": 21.8342,
        "ttft_s": 0.615,
        "prefill_tps_4k": 365.6299,
        "prefill_tps_12k": 330.4931,
        "load_seconds": 70.0669,
        "size_gb": 22.6369,
        "vram_gb": 22.6369,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "nousresearch-hermes-3-llama-3-1-8b",
      "display_name": "Hermes-3-Llama-3.1-8B",
      "model_version_id": 29,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.03,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 67.6126,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 8,
      "concurrency_text": "4 to 8 at a time",
      "perf": {
        "decode_tps": 9.5839,
        "decode_tps_c4": 21.2588,
        "ttft_s": 0.4468,
        "prefill_tps_4k": 275.7762,
        "prefill_tps_12k": 229.4447,
        "load_seconds": 72.9935,
        "size_gb": 13.7275,
        "vram_gb": 13.7275,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "internscience-sage-a1-5-q4-k-m-latest",
      "display_name": "Sage-A1.5-Q4_K_M",
      "model_version_id": 212,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 36.0,
      "size_class": "gt35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 94.4719,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 22.5502,
        "decode_tps_c4": 21.2185,
        "ttft_s": 0.5693,
        "prefill_tps_4k": 395.0861,
        "prefill_tps_12k": 374.9531,
        "load_seconds": 25.1765,
        "size_gb": 21.783,
        "vram_gb": 21.783,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "falcon3-10b-instruct",
      "display_name": "Falcon3 10B Instruct",
      "model_version_id": 96,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 10.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": null,
      "ranked": false,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 9.1919,
        "decode_tps_c4": 21.1544,
        "ttft_s": 0.518,
        "prefill_tps_4k": 176.6338,
        "prefill_tps_12k": 154.0211,
        "load_seconds": 142.203,
        "size_gb": 28.2652,
        "vram_gb": 28.2652,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "falcon3-10b-instruct",
      "display_name": "Falcon3 10B Instruct",
      "model_version_id": 189,
      "run_id": "r20261028-bv1-7ce712",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 10.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 81.2578,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 9.0354,
        "decode_tps_c4": 20.8677,
        "ttft_s": 0.5302,
        "prefill_tps_4k": 199.8057,
        "prefill_tps_12k": 173.6875,
        "load_seconds": 146.084,
        "size_gb": 6.2875,
        "vram_gb": 6.2875,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-ai-deepseek-r1-distill-qwen-7b",
      "display_name": "DeepSeek-R1-Distill-Qwen-7B",
      "model_version_id": 233,
      "run_id": "r20261105-bv1-2abe78",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.62,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 73.6455,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 13.8388,
        "decode_tps_c4": 20.6627,
        "ttft_s": 0.3101,
        "prefill_tps_4k": 341.5158,
        "prefill_tps_12k": 301.1161,
        "load_seconds": 54.1717,
        "size_gb": 4.6831,
        "vram_gb": 4.6831,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "google-gemma-3n-e4b-it",
      "display_name": "Gemma 3n E4B IT",
      "model_version_id": 9,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 86.1364,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 12.0698,
        "decode_tps_c4": 20.0344,
        "ttft_s": 0.5395,
        "prefill_tps_4k": 282.0702,
        "prefill_tps_12k": 276.8762,
        "load_seconds": 22.3761,
        "size_gb": 4.2095,
        "vram_gb": 4.2095,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-vl-30b-a3b-instruct",
      "display_name": "Qwen3-VL-30B-A3B-Instruct",
      "model_version_id": 59,
      "run_id": "r20261012-bv1-6a1291",
      "backend": "Vulkan (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 30.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.8857,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 26.9071,
        "decode_tps_c4": 19.8201,
        "ttft_s": 0.7368,
        "prefill_tps_4k": 297.6441,
        "prefill_tps_12k": 166.8251,
        "load_seconds": 12.597,
        "size_gb": 20.6197,
        "vram_gb": 20.6197,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "falcon2-11b",
      "display_name": "Falcon2 11B",
      "model_version_id": 95,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 11.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": null,
      "ranked": false,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 8.1448,
        "decode_tps_c4": 19.728,
        "ttft_s": 0.5705,
        "prefill_tps_4k": 184.0331,
        "prefill_tps_12k": 187.1082,
        "load_seconds": 31.782,
        "size_gb": 10.8793,
        "vram_gb": 10.8793,
        "gpu_fraction": 1.0,
        "context_length": 2048.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen2-math-7b-instruct",
      "display_name": "Qwen2 Math 7B Instruct",
      "model_version_id": 194,
      "run_id": "r20261028-bv1-ef0895",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": "math",
      "quantization": "Q4_K_M",
      "overall": 65.956,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 12.1515,
        "decode_tps_c4": 19.3901,
        "ttft_s": 0.3626,
        "prefill_tps_4k": 306.9087,
        "prefill_tps_12k": 319.7081,
        "load_seconds": 18.2983,
        "size_gb": 4.6831,
        "vram_gb": 4.6831,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "zai-org-glm-4-7-flash",
      "display_name": "GLM-4.7-Flash",
      "model_version_id": 64,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 31.22,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 88.9114,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 14.1416,
        "decode_tps_c4": 19.3114,
        "ttft_s": 0.5803,
        "prefill_tps_4k": 186.6383,
        "prefill_tps_12k": 104.3385,
        "load_seconds": 63.9481,
        "size_gb": 26.0261,
        "vram_gb": 26.0261,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "nemotron3-33b",
      "display_name": "Nemotron3 33B",
      "model_version_id": 226,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 33.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 86.2321,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 19.9596,
        "decode_tps_c4": 18.8633,
        "ttft_s": 0.829,
        "prefill_tps_4k": 372.0848,
        "prefill_tps_12k": 366.7239,
        "load_seconds": 24.012,
        "size_gb": 24.9648,
        "vram_gb": 24.9648,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-ai-deepseek-r1-0528-qwen3-8b",
      "display_name": "DeepSeek-R1-0528-Qwen3-8B",
      "model_version_id": 230,
      "run_id": "r20261105-bv1-2abe78",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.19,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 79.554,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 12.1974,
        "decode_tps_c4": 18.3877,
        "ttft_s": 0.3887,
        "prefill_tps_4k": 289.4775,
        "prefill_tps_12k": 229.5355,
        "load_seconds": 135.4361,
        "size_gb": 5.2254,
        "vram_gb": 5.2254,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-r1-8b-llama-distill",
      "display_name": "Deepseek R1 8B Llama Distill",
      "model_version_id": 89,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 81.7289,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 11.8841,
        "decode_tps_c4": 18.2193,
        "ttft_s": 0.3667,
        "prefill_tps_4k": 276.9593,
        "prefill_tps_12k": 230.6728,
        "load_seconds": 136.282,
        "size_gb": 22.351,
        "vram_gb": 22.351,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-6-35b-a3b",
      "display_name": "Qwen3.6-35B-A3B",
      "model_version_id": 55,
      "run_id": "r20261012-bv1-6a1291",
      "backend": "Vulkan (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 35.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 94.6282,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 22.0703,
        "decode_tps_c4": 17.8359,
        "ttft_s": 1.6289,
        "prefill_tps_4k": 276.3907,
        "prefill_tps_12k": 284.0666,
        "load_seconds": 15.1675,
        "size_gb": 23.0478,
        "vram_gb": 23.0478,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mistral-7b-instruct-v0-3",
      "display_name": "Mistral 7B Instruct V0.3",
      "model_version_id": 140,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 59.775,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 11.5781,
        "decode_tps_c4": 17.5228,
        "ttft_s": 0.4403,
        "prefill_tps_4k": 271.76,
        "prefill_tps_12k": 222.8042,
        "load_seconds": 125.86,
        "size_gb": 22.0301,
        "vram_gb": 22.0301,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen2-math-7b-instruct",
      "display_name": "Qwen2 Math 7B Instruct",
      "model_version_id": 164,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": "math",
      "quantization": "Q4_K_M",
      "overall": null,
      "ranked": false,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 11.1851,
        "decode_tps_c4": 17.4345,
        "ttft_s": 0.3723,
        "prefill_tps_4k": 309.4109,
        "prefill_tps_12k": 313.4693,
        "load_seconds": 21.859,
        "size_gb": 6.5495,
        "vram_gb": 6.5495,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "nemotron-3-5-lightning-30b-a3b",
      "display_name": "Nemotron 3.5 Lightning 30B A3B",
      "model_version_id": 224,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 30.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.8242,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 17.7208,
        "decode_tps_c4": 17.3384,
        "ttft_s": 0.8495,
        "prefill_tps_4k": 352.192,
        "prefill_tps_12k": 351.5594,
        "load_seconds": 22.6022,
        "size_gb": 25.4364,
        "vram_gb": 25.4364,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "nemotron-cascade-2-30b-a3b",
      "display_name": "Nemotron Cascade 2 30B A3B",
      "model_version_id": 225,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 30.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 85.5987,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 20.465,
        "decode_tps_c4": 17.0905,
        "ttft_s": 0.816,
        "prefill_tps_4k": 372.0278,
        "prefill_tps_12k": 367.8146,
        "load_seconds": 22.3707,
        "size_gb": 24.4388,
        "vram_gb": 24.4388,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "nvidia-nemotron-3-nano-30b-a3b",
      "display_name": "NVIDIA Nemotron 3 Nano 30B-A3B",
      "model_version_id": 30,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 30.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 83.8913,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 18.3536,
        "decode_tps_c4": 17.0396,
        "ttft_s": 0.7993,
        "prefill_tps_4k": 348.8265,
        "prefill_tps_12k": 346.3842,
        "load_seconds": 22.1702,
        "size_gb": 24.4226,
        "vram_gb": 24.4226,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "phi3-14b-medium-128k-instruct",
      "display_name": "Phi3 14B Medium 128K Instruct",
      "model_version_id": 149,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 14.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 77.9607,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": null,
        "decode_tps_c4": 17.0354,
        "ttft_s": null,
        "prefill_tps_4k": null,
        "prefill_tps_12k": 126.4916,
        "load_seconds": null,
        "size_gb": null,
        "vram_gb": null,
        "gpu_fraction": null,
        "context_length": null
      },
      "perf_units": {
        "decode_tps_c4": "tok/s",
        "prefill_tps_12k": "tok/s"
      }
    },
    {
      "slug": "olmo2-7b-1124-instruct",
      "display_name": "Olmo2 7B 1124 Instruct",
      "model_version_id": 144,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 65.361,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 11.7869,
        "decode_tps_c4": 16.8863,
        "ttft_s": 0.3674,
        "prefill_tps_4k": 260.5109,
        "prefill_tps_12k": 289.2235,
        "load_seconds": 112.265,
        "size_gb": 21.6279,
        "vram_gb": 21.6279,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mathstral-7b-v0-1",
      "display_name": "Mathstral 7B V0.1",
      "model_version_id": 136,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": "Mistral 7B",
      "specialised": "math",
      "quantization": "Q4_K_M",
      "overall": 67.987,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 12.0406,
        "decode_tps_c4": 16.7825,
        "ttft_s": 0.4174,
        "prefill_tps_4k": 271.6139,
        "prefill_tps_12k": 223.6423,
        "load_seconds": 126.843,
        "size_gb": 22.0301,
        "vram_gb": 22.0301,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-vl-4b-instruct",
      "display_name": "Qwen3-VL-4B-Instruct",
      "model_version_id": 61,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 4.44,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 89.8952,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 17.3327,
        "decode_tps_c4": 16.775,
        "ttft_s": 0.2484,
        "prefill_tps_4k": 455.209,
        "prefill_tps_12k": 363.7351,
        "load_seconds": 23.8218,
        "size_gb": 5.4601,
        "vram_gb": 5.4601,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-5-35b-a3b",
      "display_name": "Qwen3.5-35B-A3B",
      "model_version_id": 51,
      "run_id": "r20261012-bv1-6a1291",
      "backend": "Vulkan (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 35.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 94.8139,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 23.1804,
        "decode_tps_c4": 16.4638,
        "ttft_s": 1.4125,
        "prefill_tps_4k": 235.0194,
        "prefill_tps_12k": 226.924,
        "load_seconds": 15.2613,
        "size_gb": 21.7179,
        "vram_gb": 21.7179,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "aya-8b-23",
      "display_name": "Aya 8B 23",
      "model_version_id": 70,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "F16",
      "overall": 55.9173,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.9136,
        "decode_tps_c4": 16.4003,
        "ttft_s": 0.4043,
        "prefill_tps_4k": 243.0161,
        "prefill_tps_12k": 245.7259,
        "load_seconds": 84.516,
        "size_gb": 14.1562,
        "vram_gb": 14.1562,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "llama2-13b-chat",
      "display_name": "Llama2 13B Chat",
      "model_version_id": 133,
      "run_id": "r20261022-bv1-54a19b",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 13.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 45.302,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 7.0661,
        "decode_tps_c4": 16.3759,
        "ttft_s": 0.6927,
        "prefill_tps_4k": 147.5167,
        "prefill_tps_12k": 159.6716,
        "load_seconds": 109.25,
        "size_gb": 21.457,
        "vram_gb": 21.457,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "aya-expanse-8b",
      "display_name": "Aya Expanse 8B",
      "model_version_id": 71,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 72.5476,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.6996,
        "decode_tps_c4": 16.2824,
        "ttft_s": 0.4011,
        "prefill_tps_4k": 243.9431,
        "prefill_tps_12k": 243.1439,
        "load_seconds": 82.187,
        "size_gb": 14.1562,
        "vram_gb": 14.1562,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "codegeex4-9b-all",
      "display_name": "Codegeex4 9B All",
      "model_version_id": 72,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 9.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 68.678,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.0521,
        "decode_tps_c4": 16.2661,
        "ttft_s": 0.4071,
        "prefill_tps_4k": 261.6387,
        "prefill_tps_12k": 229.017,
        "load_seconds": 40.281,
        "size_gb": 11.8024,
        "vram_gb": 11.8024,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "opencoder-1-5b-instruct",
      "display_name": "Opencoder 1.5B Instruct",
      "model_version_id": 145,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 1.5,
      "size_class": "tiny",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 31.9417,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 17.4197,
        "decode_tps_c4": 16.2125,
        "ttft_s": 0.2167,
        "prefill_tps_4k": 361.0939,
        "prefill_tps_12k": 495.7045,
        "load_seconds": 47.3554,
        "size_gb": 8.4254,
        "vram_gb": 8.4254,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "codellama-13b-instruct",
      "display_name": "Codellama 13B Instruct",
      "model_version_id": 75,
      "run_id": "r20261022-bv1-54a19b",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 13.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 52.4674,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 6.9338,
        "decode_tps_c4": 16.1818,
        "ttft_s": 0.7033,
        "prefill_tps_4k": 121.7204,
        "prefill_tps_12k": 97.7222,
        "load_seconds": 444.437,
        "size_gb": 61.7475,
        "vram_gb": 61.7475,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "command-r7b-arabic-7b-02-2025",
      "display_name": "Command R7B Arabic 7B 02 2025",
      "model_version_id": 81,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 81.9195,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.9617,
        "decode_tps_c4": 16.1736,
        "ttft_s": 0.5112,
        "prefill_tps_4k": 248.2878,
        "prefill_tps_12k": 229.1643,
        "load_seconds": 92.39,
        "size_gb": 10.4904,
        "vram_gb": 10.4904,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-vl-7b-instruct",
      "display_name": "Qwen2-VL-7B-Instruct",
      "model_version_id": 43,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.29,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 72.242,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.1797,
        "decode_tps_c4": 16.1233,
        "ttft_s": 0.4258,
        "prefill_tps_4k": 331.1762,
        "prefill_tps_12k": 294.2369,
        "load_seconds": 55.3998,
        "size_gb": 12.7876,
        "vram_gb": 12.7876,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite4-1-8b",
      "display_name": "Granite4.1 8B",
      "model_version_id": 122,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.1412,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.436,
        "decode_tps_c4": 15.9417,
        "ttft_s": 0.5004,
        "prefill_tps_4k": 224.8955,
        "prefill_tps_12k": 183.7273,
        "load_seconds": 189.687,
        "size_gb": 27.1039,
        "vram_gb": 27.1039,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "glm4-9b-chat",
      "display_name": "Glm4 9B Chat",
      "model_version_id": 107,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 9.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 73.3139,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 9.8317,
        "decode_tps_c4": 15.9224,
        "ttft_s": 0.4128,
        "prefill_tps_4k": 261.2781,
        "prefill_tps_12k": 228.4236,
        "load_seconds": 40.578,
        "size_gb": 11.8024,
        "vram_gb": 11.8024,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-5-coder-14b-instruct",
      "display_name": "Qwen2.5-Coder-14B-Instruct",
      "model_version_id": 36,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 14.77,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 87.7494,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 5.1637,
        "decode_tps_c4": 15.8646,
        "ttft_s": 0.741,
        "prefill_tps_4k": 112.1146,
        "prefill_tps_12k": 86.4847,
        "load_seconds": 203.3631,
        "size_gb": 34.8467,
        "vram_gb": 34.8467,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "openai-gpt-oss-120b",
      "display_name": "gpt-oss-120b",
      "model_version_id": 67,
      "run_id": "r20261012-bv1-6a1291",
      "backend": "Vulkan (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 116.83,
      "size_class": "gt35",
      "derived_from": null,
      "specialised": null,
      "quantization": "MXFP4",
      "overall": 98.606,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 11.0328,
        "decode_tps_c4": 15.8198,
        "ttft_s": 2.7321,
        "prefill_tps_4k": 157.7708,
        "prefill_tps_12k": 140.1032,
        "load_seconds": 104.759,
        "size_gb": 64.6046,
        "vram_gb": 64.6046,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "olmo2-13b-1124-instruct",
      "display_name": "Olmo2 13B 1124 Instruct",
      "model_version_id": 143,
      "run_id": "r20261022-bv1-54a19b",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 13.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 69.5903,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 6.8386,
        "decode_tps_c4": 15.4968,
        "ttft_s": 0.6257,
        "prefill_tps_4k": 149.2482,
        "prefill_tps_12k": 161.2257,
        "load_seconds": 109.453,
        "size_gb": 21.7457,
        "vram_gb": 21.7457,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "gemma-7b-instruct-v1-1",
      "display_name": "Gemma 7B Instruct V1.1",
      "model_version_id": 101,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 7.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 60.8402,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.1814,
        "decode_tps_c4": 15.1985,
        "ttft_s": 0.4962,
        "prefill_tps_4k": 197.8626,
        "prefill_tps_12k": 199.62,
        "load_seconds": 226.094,
        "size_gb": 36.1267,
        "vram_gb": 36.1267,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite-code-8b-instruct",
      "display_name": "Granite Code 8B Instruct",
      "model_version_id": 109,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 40.6686,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.4679,
        "decode_tps_c4": 15.0173,
        "ttft_s": 0.4725,
        "prefill_tps_4k": 221.6899,
        "prefill_tps_12k": 226.6381,
        "load_seconds": 41.422,
        "size_gb": 9.9646,
        "vram_gb": 9.9646,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mistralai-ministral-3-14b-instruct-2512",
      "display_name": "Ministral 3 14B Instruct (2512)",
      "model_version_id": 26,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 14.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.7343,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 5.6275,
        "decode_tps_c4": 15.0083,
        "ttft_s": 0.7308,
        "prefill_tps_4k": 177.7442,
        "prefill_tps_12k": 134.7123,
        "load_seconds": 173.0056,
        "size_gb": 29.9808,
        "vram_gb": 29.9808,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-coder-next",
      "display_name": "Qwen3-Coder-Next",
      "model_version_id": 68,
      "run_id": "r20261012-bv1-6a1291",
      "backend": "Vulkan (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 79.67,
      "size_class": "gt35",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 96.3845,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 18.1009,
        "decode_tps_c4": 14.6929,
        "ttft_s": 1.9263,
        "prefill_tps_4k": 177.2516,
        "prefill_tps_12k": 179.5863,
        "load_seconds": 75.7438,
        "size_gb": 52.2472,
        "vram_gb": 52.2472,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-5-4b",
      "display_name": "Qwen3.5-4B",
      "model_version_id": 52,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 4.66,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 88.7306,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 15.0613,
        "decode_tps_c4": 14.6499,
        "ttft_s": 0.4221,
        "prefill_tps_4k": 359.9639,
        "prefill_tps_12k": 347.9449,
        "load_seconds": 11.0953,
        "size_gb": 3.6291,
        "vram_gb": 3.6291,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite3-dense-8b-instruct",
      "display_name": "Granite3 Dense 8B Instruct",
      "model_version_id": 119,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 67.564,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 10.4557,
        "decode_tps_c4": 14.4574,
        "ttft_s": 0.4866,
        "prefill_tps_4k": 255.2237,
        "prefill_tps_12k": 256.7021,
        "load_seconds": 46.64,
        "size_gb": 10.5426,
        "vram_gb": 10.5426,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-8b",
      "display_name": "Qwen3-8B",
      "model_version_id": 57,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.19,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 91.5462,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 9.2875,
        "decode_tps_c4": 14.3728,
        "ttft_s": 0.4989,
        "prefill_tps_4k": 249.4424,
        "prefill_tps_12k": 202.9387,
        "load_seconds": 148.5704,
        "size_gb": 24.6665,
        "vram_gb": 24.6665,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "coherelabs-c4ai-command-r7b-12-2024",
      "display_name": "Command R7B (12-2024)",
      "model_version_id": 2,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 74.798,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 8.6634,
        "decode_tps_c4": 14.0239,
        "ttft_s": 0.557,
        "prefill_tps_4k": 265.3425,
        "prefill_tps_12k": 266.0527,
        "load_seconds": 60.2003,
        "size_gb": 10.4233,
        "vram_gb": 10.4233,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-14b-chat-v1-5",
      "display_name": "Qwen 14B Chat V1.5",
      "model_version_id": 154,
      "run_id": "r20261022-bv1-54a19b",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 14.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 54.7805,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 6.2729,
        "decode_tps_c4": 13.9018,
        "ttft_s": 0.7251,
        "prefill_tps_4k": 136.4316,
        "prefill_tps_12k": 104.5296,
        "load_seconds": 462.547,
        "size_gb": 62.5771,
        "vram_gb": 62.5771,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "solar-10-7b-instruct-v1",
      "display_name": "Solar 10.7B Instruct V1",
      "model_version_id": 177,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 10.7,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 58.6833,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 8.6758,
        "decode_tps_c4": 13.5854,
        "ttft_s": 0.609,
        "prefill_tps_4k": 182.8502,
        "prefill_tps_12k": 190.5144,
        "load_seconds": 47.25,
        "size_gb": 13.0813,
        "vram_gb": 13.0813,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "aya-expanse-32b",
      "display_name": "Aya Expanse 32B",
      "model_version_id": 196,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 32.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 78.7594,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 9.9938,
        "decode_tps_c4": 13.5274,
        "ttft_s": 0.4201,
        "prefill_tps_4k": 281.2772,
        "prefill_tps_12k": 88.1637,
        "load_seconds": 89.1578,
        "size_gb": 31.4334,
        "vram_gb": 31.4334,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "google-gemma-4-12b-it",
      "display_name": "Gemma 4 12B IT",
      "model_version_id": 10,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 12.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 98.5523,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 8.119,
        "decode_tps_c4": 13.1275,
        "ttft_s": 0.9368,
        "prefill_tps_4k": 156.8684,
        "prefill_tps_12k": 147.8502,
        "load_seconds": 89.0233,
        "size_gb": 8.9078,
        "vram_gb": 8.9078,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "nemotron-3-5-bolt-e4b",
      "display_name": "Nemotron 3.5 Bolt E4B",
      "model_version_id": 223,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 90.2903,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 9.7681,
        "decode_tps_c4": 12.995,
        "ttft_s": 0.3939,
        "prefill_tps_4k": 383.7537,
        "prefill_tps_12k": 364.5984,
        "load_seconds": 41.6602,
        "size_gb": 5.7356,
        "vram_gb": 5.7356,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mistral-nemo-12b-instruct-2407",
      "display_name": "Mistral Nemo 12B Instruct 2407",
      "model_version_id": 141,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 12.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 75.9002,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 8.1804,
        "decode_tps_c4": 12.5021,
        "ttft_s": 0.5436,
        "prefill_tps_4k": 176.6196,
        "prefill_tps_12k": 152.6864,
        "load_seconds": 143.516,
        "size_gb": 29.1539,
        "vram_gb": 29.1539,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite4-small-h",
      "display_name": "Granite4 Small H",
      "model_version_id": 211,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 32.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 93.9127,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 8.4433,
        "decode_tps_c4": 12.3916,
        "ttft_s": 1.4353,
        "prefill_tps_4k": 133.1758,
        "prefill_tps_12k": 135.4183,
        "load_seconds": 37.9266,
        "size_gb": 23.9189,
        "vram_gb": 23.9189,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "google-gemma-4-31b-it",
      "display_name": "gemma-4-31B-it",
      "model_version_id": 12,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 31.27,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 99.3091,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 3.8389,
        "decode_tps_c4": 11.7935,
        "ttft_s": 2.3682,
        "prefill_tps_4k": 63.6962,
        "prefill_tps_12k": 57.5627,
        "load_seconds": 122.6714,
        "size_gb": 11.1594,
        "vram_gb": 11.1594,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "google-gemma-2-9b-it",
      "display_name": "gemma-2-9b-it",
      "model_version_id": 5,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 9.24,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 79.8919,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 7.5773,
        "decode_tps_c4": 11.7128,
        "ttft_s": 0.6862,
        "prefill_tps_4k": 171.6401,
        "prefill_tps_12k": 171.3629,
        "load_seconds": 154.3646,
        "size_gb": 14.7141,
        "vram_gb": 14.7141,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "translategemma-12b-it",
      "display_name": "Translategemma 12B IT",
      "model_version_id": 182,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 12.0,
      "size_class": "9-16",
      "derived_from": "Gemma 3",
      "specialised": "translation",
      "quantization": "Q4_K_M",
      "overall": 89.6385,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 7.6397,
        "decode_tps_c4": 11.538,
        "ttft_s": 0.7732,
        "prefill_tps_4k": 176.8125,
        "prefill_tps_12k": 175.8798,
        "load_seconds": 97.297,
        "size_gb": 13.0755,
        "vram_gb": 13.0755,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mistral-small-24b-instruct-2501",
      "display_name": "Mistral Small 24B Instruct 2501",
      "model_version_id": 219,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 24.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.138,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 4.2181,
        "decode_tps_c4": 10.5787,
        "ttft_s": 1.032,
        "prefill_tps_4k": 98.4382,
        "prefill_tps_12k": 87.8892,
        "load_seconds": 204.3266,
        "size_gb": 36.4636,
        "vram_gb": 36.4636,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "ornith-9b",
      "display_name": "Ornith 9B",
      "model_version_id": 147,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 9.0,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 89.9782,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 10.7962,
        "decode_tps_c4": 10.5076,
        "ttft_s": 0.797,
        "prefill_tps_4k": 261.6314,
        "prefill_tps_12k": 256.3338,
        "load_seconds": 12.312,
        "size_gb": 5.8873,
        "vram_gb": 5.8873,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mistralai-magistral-small-2506",
      "display_name": "Magistral Small (2506)",
      "model_version_id": 236,
      "run_id": "r20261105-bv1-2abe78",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 24.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 49.7972,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 4.5266,
        "decode_tps_c4": 10.4745,
        "ttft_s": 1.0413,
        "prefill_tps_4k": 106.8456,
        "prefill_tps_12k": 94.8824,
        "load_seconds": 153.8431,
        "size_gb": 14.3339,
        "vram_gb": 14.3339,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepcoder-14b-preview",
      "display_name": "Deepcoder 14B Preview",
      "model_version_id": 83,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 14.0,
      "size_class": "9-16",
      "derived_from": "DeepSeek-R1-Distill-Qwen 14B",
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 74.4947,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 6.6966,
        "decode_tps_c4": 10.3271,
        "ttft_s": 0.6709,
        "prefill_tps_4k": 145.7093,
        "prefill_tps_12k": 107.9164,
        "load_seconds": 198.953,
        "size_gb": 34.8887,
        "vram_gb": 34.8887,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mistral-small3-1-24b-instruct-2503",
      "display_name": "Mistral Small3.1 24B Instruct 2503",
      "model_version_id": 220,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 24.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 90.1023,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 4.3107,
        "decode_tps_c4": 10.2841,
        "ttft_s": 1.3437,
        "prefill_tps_4k": 78.7303,
        "prefill_tps_12k": 75.4475,
        "load_seconds": 154.515,
        "size_gb": 36.6328,
        "vram_gb": 36.6328,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "cogito-14b-v1-preview-qwen",
      "display_name": "Cogito 14B V1 Preview Qwen",
      "model_version_id": 78,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 14.0,
      "size_class": "9-16",
      "derived_from": "Qwen 2.5",
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 90.2542,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 6.6269,
        "decode_tps_c4": 10.208,
        "ttft_s": 0.7463,
        "prefill_tps_4k": 140.3235,
        "prefill_tps_12k": 105.7577,
        "load_seconds": 200.328,
        "size_gb": 34.887,
        "vram_gb": 34.887,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-ai-deepseek-r1-distill-qwen-14b",
      "display_name": "DeepSeek-R1-Distill-Qwen-14B",
      "model_version_id": 231,
      "run_id": "r20261105-bv1-2abe78",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 14.77,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 85.2897,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 7.1022,
        "decode_tps_c4": 9.9981,
        "ttft_s": 0.6174,
        "prefill_tps_4k": 146.8286,
        "prefill_tps_12k": 104.4517,
        "load_seconds": 173.3613,
        "size_gb": 8.9881,
        "vram_gb": 8.9881,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "google-gemma-3-12b-it",
      "display_name": "gemma-3-12b-it",
      "model_version_id": 6,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 12.19,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.742,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 6.1155,
        "decode_tps_c4": 9.4774,
        "ttft_s": 0.843,
        "prefill_tps_4k": 139.0193,
        "prefill_tps_12k": 138.4609,
        "load_seconds": 115.3567,
        "size_gb": 13.0607,
        "vram_gb": 13.0607,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mistralai-mistral-small-3-2-24b-instruct-2506",
      "display_name": "Mistral Small 3.2 24B Instruct (2506)",
      "model_version_id": 28,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 24.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.8816,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.1182,
        "decode_tps_c4": 9.3526,
        "ttft_s": 1.7174,
        "prefill_tps_4k": 98.934,
        "prefill_tps_12k": 79.6846,
        "load_seconds": 176.9068,
        "size_gb": 36.4783,
        "vram_gb": 36.4783,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-vl-8b-instruct",
      "display_name": "Qwen3-VL-8B-Instruct",
      "model_version_id": 62,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.77,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 94.8954,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 9.3756,
        "decode_tps_c4": 9.1811,
        "ttft_s": 0.4322,
        "prefill_tps_4k": 261.4355,
        "prefill_tps_12k": 228.9543,
        "load_seconds": 25.868,
        "size_gb": 7.7609,
        "vram_gb": 7.7609,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "microsoft-phi-4",
      "display_name": "phi-4",
      "model_version_id": 23,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 14.66,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 85.9357,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 4.9913,
        "decode_tps_c4": 8.9447,
        "ttft_s": 0.7453,
        "prefill_tps_4k": 135.446,
        "prefill_tps_12k": 122.4466,
        "load_seconds": 210.5625,
        "size_gb": 36.1955,
        "vram_gb": 36.1955,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mistralai-devstral-small-2-24b-instruct-2512",
      "display_name": "Devstral Small 2 24B Instruct (2512)",
      "model_version_id": 25,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 24.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 92.0736,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.1363,
        "decode_tps_c4": 8.8856,
        "ttft_s": 1.5395,
        "prefill_tps_4k": 79.9954,
        "prefill_tps_12k": 64.8535,
        "load_seconds": 177.645,
        "size_gb": 36.4783,
        "vram_gb": 36.4783,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "muse-glimmer-30b",
      "display_name": "Muse Glimmer 30B",
      "model_version_id": 222,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 30.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 97.8982,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.6388,
        "decode_tps_c4": 8.8104,
        "ttft_s": 2.1202,
        "prefill_tps_4k": 96.0664,
        "prefill_tps_12k": 96.4292,
        "load_seconds": 42.2923,
        "size_gb": 18.329,
        "vram_gb": 18.329,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "swiss-ai-apertus-8b-instruct-2509",
      "display_name": "Apertus-8B-Instruct-2509",
      "model_version_id": 63,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 8.05,
      "size_class": "4-9",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": null,
      "ranked": false,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 9.3723,
        "decode_tps_c4": 8.7309,
        "ttft_s": 0.4224,
        "prefill_tps_4k": 293.8039,
        "prefill_tps_12k": 243.2448,
        "load_seconds": 115.3286,
        "size_gb": 22.2718,
        "vram_gb": 22.2718,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-14b",
      "display_name": "Qwen3-14B",
      "model_version_id": 44,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 14.77,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 91.707,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 4.9961,
        "decode_tps_c4": 8.6657,
        "ttft_s": 0.8052,
        "prefill_tps_4k": 153.4525,
        "prefill_tps_12k": 112.9015,
        "load_seconds": 171.4608,
        "size_gb": 30.9283,
        "vram_gb": 30.9283,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mixtral-8x7b-instruct-v0-1",
      "display_name": "Mixtral 8x7B Instruct V0.1",
      "model_version_id": 221,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 56.0,
      "size_class": "gt35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 56.7665,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 7.5847,
        "decode_tps_c4": 8.494,
        "ttft_s": 1.3171,
        "prefill_tps_4k": 136.8653,
        "prefill_tps_12k": 127.0171,
        "load_seconds": 139.9958,
        "size_gb": 46.4599,
        "vram_gb": 46.4599,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-5-9b",
      "display_name": "Qwen3.5-9B",
      "model_version_id": 53,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 9.65,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.039,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 8.4176,
        "decode_tps_c4": 8.2368,
        "ttft_s": 0.6868,
        "prefill_tps_4k": 225.4248,
        "prefill_tps_12k": 221.6891,
        "load_seconds": 13.5315,
        "size_gb": 6.1139,
        "vram_gb": 6.1139,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "meta-llama-llama-4-scout-17b-16e-instruct",
      "display_name": "Llama 4 Scout 17B-16E Instruct",
      "model_version_id": 66,
      "run_id": "r20261012-bv1-6a1291",
      "backend": "Vulkan (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 109.0,
      "size_class": "gt35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 93.5581,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 16.9535,
        "decode_tps_c4": 8.2324,
        "ttft_s": 2.5258,
        "prefill_tps_4k": 147.436,
        "prefill_tps_12k": 162.8656,
        "load_seconds": 81.4357,
        "size_gb": 70.3669,
        "vram_gb": 70.3669,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite4-2-30b",
      "display_name": "Granite4.2 30B",
      "model_version_id": 208,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 30.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.8069,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.3926,
        "decode_tps_c4": 8.2145,
        "ttft_s": 1.3651,
        "prefill_tps_4k": 62.0325,
        "prefill_tps_12k": 59.6169,
        "load_seconds": 243.2704,
        "size_gb": 52.3497,
        "vram_gb": 52.3497,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "devstral-24b-small-2505",
      "display_name": "Devstral 24B Small 2505",
      "model_version_id": 203,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 24.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 88.945,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 4.1843,
        "decode_tps_c4": 8.0061,
        "ttft_s": 1.5394,
        "prefill_tps_4k": 121.9437,
        "prefill_tps_12k": 86.047,
        "load_seconds": 165.404,
        "size_gb": 36.4636,
        "vram_gb": 36.4636,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "ornith-ai-ornith-1-5-9b",
      "display_name": "Ornith-1.5-9B",
      "model_version_id": 31,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 9.65,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 88.9704,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 8.1416,
        "decode_tps_c4": 7.9859,
        "ttft_s": 0.7055,
        "prefill_tps_4k": 219.6858,
        "prefill_tps_12k": 214.817,
        "load_seconds": 11.6635,
        "size_gb": 6.3955,
        "vram_gb": 6.3955,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "command-r-35b-08-2024",
      "display_name": "Command R 35B 08 2024",
      "model_version_id": 201,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 35.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 75.1743,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.2315,
        "decode_tps_c4": 7.6874,
        "ttft_s": 1.3449,
        "prefill_tps_4k": 60.9149,
        "prefill_tps_12k": 55.0572,
        "load_seconds": 158.9786,
        "size_gb": 42.2044,
        "vram_gb": 42.2044,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "exaone3-5-32b-instruct",
      "display_name": "Exaone3.5 32B Instruct",
      "model_version_id": 204,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 32.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 85.0293,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.1191,
        "decode_tps_c4": 7.6686,
        "ttft_s": 1.3083,
        "prefill_tps_4k": 66.3634,
        "prefill_tps_12k": 56.7781,
        "load_seconds": 256.6016,
        "size_gb": 53.8584,
        "vram_gb": 53.8584,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-5-coder-32b-instruct",
      "display_name": "Qwen2.5-Coder-32B-Instruct",
      "model_version_id": 37,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 32.76,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 90.5357,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 2.3009,
        "decode_tps_c4": 7.6424,
        "ttft_s": 1.6116,
        "prefill_tps_4k": 51.762,
        "prefill_tps_12k": 45.8706,
        "load_seconds": 244.5076,
        "size_gb": 54.2034,
        "vram_gb": 54.2034,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-5-32b-instruct",
      "display_name": "Qwen2.5-32B-Instruct",
      "model_version_id": 33,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 32.76,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 93.5071,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 2.3043,
        "decode_tps_c4": 7.6125,
        "ttft_s": 1.6149,
        "prefill_tps_4k": 52.5156,
        "prefill_tps_12k": 46.1117,
        "load_seconds": 251.6695,
        "size_gb": 54.2034,
        "vram_gb": 54.2034,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "zai-org-glm-4-5-air",
      "display_name": "GLM-4.5-Air",
      "model_version_id": 69,
      "run_id": "r20261015-bv1-5ac997",
      "backend": "Vulkan (Ollama 0.40.4, flash attention, 16k context; 2 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 110.47,
      "size_class": "gt35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.7568,
      "ranked": true,
      "concurrency_min": 2,
      "concurrency_max": 2,
      "concurrency_text": "2 at a time",
      "perf": {
        "decode_tps": 5.9864,
        "decode_tps_c4": 7.5927,
        "ttft_s": 3.1499,
        "prefill_tps_4k": 53.4418,
        "prefill_tps_12k": 28.8813,
        "load_seconds": 110.5245,
        "size_gb": 76.7538,
        "vram_gb": 76.7538,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "bartowski-nanbeige-nanbeige4-3-33b",
      "display_name": "Nanbeige_Nanbeige4.3-33B",
      "model_version_id": 197,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 33.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 90.3666,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.0976,
        "decode_tps_c4": 7.3638,
        "ttft_s": 1.5413,
        "prefill_tps_4k": 70.8218,
        "prefill_tps_12k": 58.1961,
        "load_seconds": 200.9046,
        "size_gb": 48.8722,
        "vram_gb": 48.8722,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "codellama-34b-instruct",
      "display_name": "Codellama 34B Instruct",
      "model_version_id": 199,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 34.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 61.217,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.1316,
        "decode_tps_c4": 7.3637,
        "ttft_s": 1.6169,
        "prefill_tps_4k": 58.0542,
        "prefill_tps_12k": 50.7601,
        "load_seconds": 190.1519,
        "size_gb": 46.7182,
        "vram_gb": 46.7182,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "olmo-3-1-32b-instruct",
      "display_name": "Olmo 3.1 32B Instruct",
      "model_version_id": 227,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 32.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 89.5821,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.1416,
        "decode_tps_c4": 7.3567,
        "ttft_s": 2.6116,
        "prefill_tps_4k": 55.5619,
        "prefill_tps_12k": 53.7029,
        "load_seconds": 128.4439,
        "size_gb": 27.7104,
        "vram_gb": 27.7104,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "dphn-dolphin-2-9-1-yi-1-5-34b",
      "display_name": "dolphin-2.9.1-yi-1.5-34b",
      "model_version_id": 4,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 34.39,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 72.964,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 2.4005,
        "decode_tps_c4": 7.191,
        "ttft_s": 1.9447,
        "prefill_tps_4k": 50.7374,
        "prefill_tps_12k": 51.347,
        "load_seconds": 123.0628,
        "size_gb": 37.2184,
        "vram_gb": 37.2184,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "yi-34b-chat-v1-5",
      "display_name": "YI 34B Chat V1.5",
      "model_version_id": 228,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 34.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 75.2062,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.004,
        "decode_tps_c4": 7.1484,
        "ttft_s": 1.6171,
        "prefill_tps_4k": 56.8029,
        "prefill_tps_12k": 59.1703,
        "load_seconds": 71.6967,
        "size_gb": 28.8297,
        "vram_gb": 28.8297,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "stablelm2-12b-chat",
      "display_name": "Stablelm2 12B Chat",
      "model_version_id": 181,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 12.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 42.4203,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 4.737,
        "decode_tps_c4": 6.7286,
        "ttft_s": 0.8462,
        "prefill_tps_4k": 84.0702,
        "prefill_tps_12k": 102.7095,
        "load_seconds": 50.187,
        "size_gb": 14.0397,
        "vram_gb": 14.0397,
        "gpu_fraction": 1.0,
        "context_length": 4096.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "codestral-22b-v0-1",
      "display_name": "Codestral 22B V0.1",
      "model_version_id": 200,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 22.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 74.7525,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 4.4313,
        "decode_tps_c4": 6.5467,
        "ttft_s": 1.1655,
        "prefill_tps_4k": 76.1053,
        "prefill_tps_12k": 64.4064,
        "load_seconds": 241.9736,
        "size_gb": 43.9631,
        "vram_gb": 43.9631,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "mistral-small-22b-instruct-2409",
      "display_name": "Mistral Small 22B Instruct 2409",
      "model_version_id": 218,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 22.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 84.6347,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 4.2313,
        "decode_tps_c4": 6.1178,
        "ttft_s": 1.1861,
        "prefill_tps_4k": 76.2456,
        "prefill_tps_12k": 64.834,
        "load_seconds": 244.6143,
        "size_gb": 43.9631,
        "vram_gb": 43.9631,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-v2-16b-lite-chat",
      "display_name": "Deepseek V2 16B Lite Chat",
      "model_version_id": 90,
      "run_id": "r20261016-bv1-473145",
      "backend": "ROCm 7.1 (Ollama 0.40.4, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 16.0,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 58.1802,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 12.1741,
        "decode_tps_c4": 5.9416,
        "ttft_s": 0.4913,
        "prefill_tps_4k": 244.1408,
        "prefill_tps_12k": 108.9644,
        "load_seconds": 243.39,
        "size_gb": 46.8174,
        "vram_gb": 46.8174,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-coder-33b-instruct",
      "display_name": "Deepseek Coder 33B Instruct",
      "model_version_id": 202,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 33.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 33.4378,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.0045,
        "decode_tps_c4": 5.93,
        "ttft_s": 1.6182,
        "prefill_tps_4k": 63.6249,
        "prefill_tps_12k": 54.3164,
        "load_seconds": 239.9373,
        "size_gb": 53.4823,
        "vram_gb": 53.4823,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "gemma2-27b-instruct",
      "display_name": "Gemma2 27B Instruct",
      "model_version_id": 206,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 27.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 85.5807,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.6926,
        "decode_tps_c4": 5.5196,
        "ttft_s": 1.5213,
        "prefill_tps_4k": 67.0124,
        "prefill_tps_12k": 67.9504,
        "load_seconds": 157.6047,
        "size_gb": 27.0161,
        "vram_gb": 27.0161,
        "gpu_fraction": 1.0,
        "context_length": 8192.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "medgemma-27b-it",
      "display_name": "Medgemma 27B IT",
      "model_version_id": 217,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 27.0,
      "size_class": "16-35",
      "derived_from": "Gemma 3",
      "specialised": "medical",
      "quantization": "Q4_K_M",
      "overall": 90.2453,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.5467,
        "decode_tps_c4": 5.3201,
        "ttft_s": 1.6361,
        "prefill_tps_4k": 87.7612,
        "prefill_tps_12k": 89.0226,
        "load_seconds": 126.635,
        "size_gb": 22.0907,
        "vram_gb": 22.0907,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "granite4-1-30b",
      "display_name": "Granite4.1 30B",
      "model_version_id": 207,
      "run_id": "r20261102-bv1-5246ab",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 30.0,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.3162,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.3861,
        "decode_tps_c4": 5.0011,
        "ttft_s": 1.4861,
        "prefill_tps_4k": 62.4365,
        "prefill_tps_12k": 59.1875,
        "load_seconds": 242.0584,
        "size_gb": 52.3497,
        "vram_gb": 52.3497,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-ai-deepseek-r1-distill-qwen-32b",
      "display_name": "DeepSeek-R1-Distill-Qwen-32B",
      "model_version_id": 232,
      "run_id": "r20261105-bv1-2abe78",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 32.76,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 89.0854,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.095,
        "decode_tps_c4": 4.7697,
        "ttft_s": 1.3583,
        "prefill_tps_4k": 56.4769,
        "prefill_tps_12k": 48.8082,
        "load_seconds": 281.5239,
        "size_gb": 19.8513,
        "vram_gb": 19.8513,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwq-32b",
      "display_name": "QwQ 32B",
      "model_version_id": 239,
      "run_id": "r20261105-bv1-2abe78",
      "backend": "ROCm 7.1 (Ollama 0.40.13, flash attention, 16k context; 8 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 32.8,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 92.6424,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 3.1549,
        "decode_tps_c4": 4.7182,
        "ttft_s": 1.4579,
        "prefill_tps_4k": 72.092,
        "prefill_tps_12k": 60.1915,
        "load_seconds": 271.7463,
        "size_gb": 54.2244,
        "vram_gb": 54.2244,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "google-gemma-3-27b-it",
      "display_name": "gemma-3-27b-it",
      "model_version_id": 7,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 27.43,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 95.7673,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 2.7226,
        "decode_tps_c4": 4.6898,
        "ttft_s": 1.9135,
        "prefill_tps_4k": 70.8791,
        "prefill_tps_12k": 71.5421,
        "load_seconds": 124.9378,
        "size_gb": 22.0768,
        "vram_gb": 22.0768,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "deepseek-ai-deepseek-coder-v2-lite-instruct",
      "display_name": "DeepSeek-Coder-V2-Lite-Instruct",
      "model_version_id": 3,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 15.71,
      "size_class": "9-16",
      "derived_from": null,
      "specialised": "code",
      "quantization": "Q4_K_M",
      "overall": 78.9798,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 11.0589,
        "decode_tps_c4": 4.5617,
        "ttft_s": 0.5033,
        "prefill_tps_4k": 159.8893,
        "prefill_tps_12k": 110.0963,
        "load_seconds": 283.9216,
        "size_gb": 46.8006,
        "vram_gb": 46.8006,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-32b",
      "display_name": "Qwen3-32B",
      "model_version_id": 47,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 32.76,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 93.6304,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 2.2793,
        "decode_tps_c4": 4.1371,
        "ttft_s": 1.7561,
        "prefill_tps_4k": 54.297,
        "prefill_tps_12k": 45.5854,
        "load_seconds": 246.2199,
        "size_gb": 54.9647,
        "vram_gb": 54.9647,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-6-27b",
      "display_name": "Qwen3.6-27B",
      "model_version_id": 54,
      "run_id": "r20261012-bv1-6a1291",
      "backend": "Vulkan (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 27.78,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 95.0871,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": null,
        "decode_tps_c4": 3.4518,
        "ttft_s": null,
        "prefill_tps_4k": 55.246,
        "prefill_tps_12k": 54.5776,
        "load_seconds": 11.8451,
        "size_gb": 17.3089,
        "vram_gb": 17.3089,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "meta-llama-llama-3-3-70b-instruct",
      "display_name": "Llama-3.3-70B-Instruct",
      "model_version_id": 65,
      "run_id": "r20261012-bv1-6a1291",
      "backend": "Vulkan (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 70.55,
      "size_class": "gt35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 94.2509,
      "ranked": true,
      "concurrency_min": 4,
      "concurrency_max": 4,
      "concurrency_text": "4 at a time",
      "perf": {
        "decode_tps": 1.3738,
        "decode_tps_c4": 3.4367,
        "ttft_s": 4.3013,
        "prefill_tps_4k": 15.5233,
        "prefill_tps_12k": 11.4146,
        "load_seconds": 51.0511,
        "size_gb": 63.8494,
        "vram_gb": 63.8494,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-8-27b",
      "display_name": "Qwen3.8-27B",
      "model_version_id": 56,
      "run_id": "r20261012-bv1-6a1291",
      "backend": "Vulkan (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 27.78,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 95.4183,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 3.4517,
        "decode_tps_c4": 3.1692,
        "ttft_s": 4.1338,
        "prefill_tps_4k": 51.919,
        "prefill_tps_12k": 51.2641,
        "load_seconds": 12.2341,
        "size_gb": 17.6224,
        "vram_gb": 17.6224,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-vl-32b-instruct",
      "display_name": "Qwen3-VL-32B-Instruct",
      "model_version_id": 60,
      "run_id": "r20261012-bv1-6a1291",
      "backend": "Vulkan (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 33.36,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 94.6605,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 3.1632,
        "decode_tps_c4": 2.8913,
        "ttft_s": 1.4851,
        "prefill_tps_4k": 32.7342,
        "prefill_tps_12k": 21.6184,
        "load_seconds": 15.6193,
        "size_gb": 24.5243,
        "vram_gb": 24.5243,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen2-5-vl-32b-instruct",
      "display_name": "Qwen2.5-VL-32B-Instruct",
      "model_version_id": 40,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 33.45,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 89.7263,
      "ranked": true,
      "concurrency_min": 1,
      "concurrency_max": 1,
      "concurrency_text": "1 at a time",
      "perf": {
        "decode_tps": 2.3122,
        "decode_tps_c4": 2.6575,
        "ttft_s": 1.6133,
        "prefill_tps_4k": 51.5838,
        "prefill_tps_12k": null,
        "load_seconds": 248.3929,
        "size_gb": 54.9396,
        "vram_gb": 54.9396,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    },
    {
      "slug": "qwen-qwen3-5-27b",
      "display_name": "Qwen3.5-27B",
      "model_version_id": 50,
      "run_id": "r20261006-bv1-53e146",
      "backend": "ROCm 7.1 (Ollama 0.34.4, flash attention, 16k context; 4 parallel slots until 2026-10-07 00:45Z, 8 after)",
      "gpu": "AMD Radeon 8060S (Ryzen AI Max+ 395, 128 GB unified, ~78 GB GPU)",
      "params_b": 27.78,
      "size_class": "16-35",
      "derived_from": null,
      "specialised": null,
      "quantization": "Q4_K_M",
      "overall": 96.1888,
      "ranked": true,
      "concurrency_min": 8,
      "concurrency_max": 8,
      "concurrency_text": "8 at a time",
      "perf": {
        "decode_tps": 2.7213,
        "decode_tps_c4": 2.648,
        "ttft_s": 2.0584,
        "prefill_tps_4k": 70.5111,
        "prefill_tps_12k": 69.0165,
        "load_seconds": 27.6661,
        "size_gb": 17.6015,
        "vram_gb": 17.6015,
        "gpu_fraction": 1.0,
        "context_length": 16384.0
      },
      "perf_units": {
        "context_length": "tokens",
        "decode_tps": "tok/s",
        "decode_tps_c4": "tok/s",
        "gpu_fraction": "ratio",
        "load_seconds": "s",
        "prefill_tps_12k": "tok/s",
        "prefill_tps_4k": "tok/s",
        "size_gb": "GB",
        "ttft_s": "s",
        "vram_gb": "GB"
      }
    }
  ]
}
