What runs on my hardware
Enter the memory you can give a model, in GB. This page lists the ranked models whose estimated memory need fits, best Overall score first. The estimate is the model file size plus about 20% for the context of a single conversation. You can also set the slowest speed you would accept. Back to the leaderboard.
Read the memory figure as an estimate, not a guarantee. The memory a model needs is estimated from its file: the weights plus about 20% for the context of a single conversation. It is not a measurement, and a long conversation or several at once needs more. A model with no file size on record is not counted as fitting any amount of memory. Speeds, and the GPU memory shown on the speed page, were measured on this benchmark's server, not on yours. That server kept 1 to 8 answers in flight at a time and loaded a context window of 16,384 tokens for most models; 38 of the models listed here ran at a shorter context, shown in the Context column. Speeds were measured on one laptop and are comparable only between models on the same backend. See the settings and the memory method.
Models that fit in 16 GB
165 of 227 ranked models fit. A model that needs exactly 16 GB is included. 11 unranked models are not listed; see the leaderboard.
| Rank | Model | Size class | Overall | Decode (tok/s) | Memory needed (GB, est.) | Context (tokens) |
|---|---|---|---|---|---|---|
| 1–12 | Gemma 4 12B IT | 9–16B | 98.697.0–99.8 | 8.1ROCm 7.1 | 8.9 | 16,384 |
| 2–51 | Gemma4 E4B IT | 4–9B | 95.593.0–97.7 | 13.2ROCm 7.1 | 7.9 | 16,384 |
| 4–48 | Qwen3-VL-8B-Instruct | 4–9B | 94.992.4–97.1 | 9.4ROCm 7.1 | 7.4 | 16,384 |
| 4–61 | Gemma4 E2B IT | 4–9B | 93.891.0–96.1 | 17.2ROCm 7.1 | 5.5 | 16,384 |
| 8–66 | gemma-3-12b-it | 9–16B | 92.789.7–95.4 | 6.1ROCm 7.1 | 9.8 | 16,384 |
| 9–61 | Ministral 3 14B Instruct (2512) | 9–16B | 92.790.0–95.2 | 5.6ROCm 7.1 | 10.9 | 16,384 |
| 6–73 | Granite4.1 8B | 4–9B | 92.189.0–95.0 | 10.4ROCm 7.1 | 6.4 | 16,384 |
| 11–69 | Qwen3.5-9B | 9–16B | 92.089.0–95.1 | 8.4ROCm 7.1 | 6.8 | 16,384 |
| 11–71 | Qwen3-14B | 9–16B | 91.788.5–94.7 | 5.0ROCm 7.1 | 11.1 | 16,384 |
| 11–72 | Qwen3-8B | 4–9B | 91.588.4–94.5 | 9.3ROCm 7.1 | 6.3 | 16,384 |
| 11–72 | Qwen2.5-14B-Instruct | 9–16B | 91.388.0–94.2 | 21.7ROCm 7.1 | 10.8 | 16,384 |
| 8–82 | Nemotron 3.5 Bolt E4B | 4–9B | 90.386.3–93.9 | 9.8ROCm 7.1 | 8.6 | 16,384 |
| 13–73 | Qwen3-4B-Instruct-2507 | 4–9B | 90.387.2–93.3 | 17.1ROCm 7.1 | 3.0 | 16,384 |
| 11–79 | Cogito 14B V1 Preview Qwenfine-tune of Qwen 2.5 | 9–16B | 90.386.8–93.4 | 6.6ROCm 7.1 | 10.8 | 16,384 |
| 12–79 | Ornith 9B | 4–9B | 90.086.5–93.1 | 10.8ROCm 7.1 | 6.7 | 16,384 |
| 12–76 | Qwen3-VL-4B-Instruct | 4–9B | 89.986.4–93.1 | 17.3ROCm 7.1 | 4.0 | 16,384 |
| 13–80 | Translategemma 12B ITfine-tune of Gemma 3translation model | 9–16B | 89.686.4–92.9 | 7.6ROCm 7.1 | 9.7 | 16,384 |
| 18–81 | Qwen3-4B | 4–9B | 89.285.6–92.4 | 17.0ROCm 7.1 | 3.1 | 16,384 |
| 19–84 | Qwen3.5-4B | 4–9B | 88.784.9–92.4 | 15.1ROCm 7.1 | 3.2 | 16,384 |
| 19–86 | Ministral 3 8B Instruct 2512 | 4–9B | 88.785.2–92.1 | 11.0ROCm 7.1 | 7.2 | 16,384 |
| 30–86 | Qwen2.5-Coder-14B-Instructcode model | 9–16B | 87.784.0–91.3 | 5.2ROCm 7.1 | 10.8 | 16,384 |
| 44–97 | Gemma 3n E4B IT | 4–9B | 86.182.4–89.6 | 12.1ROCm 7.1 | 9.1 | 16,384 |
| 39–99 | phi-4 | 9–16B | 85.981.9–89.8 | 5.0ROCm 7.1 | 10.9 | 16,384 |
| 45–100 | Qwen2.5-7B-Instruct | 4–9B | 85.681.4–89.3 | 10.1ROCm 7.1 | 5.6 | 16,384 |
| 38–106 | DeepSeek-R1-Distill-Qwen-14B | 9–16B | 85.381.2–89.2 | 7.1ROCm 7.1 | 10.8 | 16,384 |
| 43–106 | Spark-X2.6-4B | 3–4B | 85.080.9–89.1 | 19.7ROCm 7.1 | 3.0 | 16,384 |
| 50–106 | Qwen2.5-Coder-7B-Instructcode model | 4–9B | 84.580.6–88.1 | 9.8ROCm 7.1 | 5.6 | 16,384 |
| 47–105 | Cogito 8B V1 Preview Llamafine-tune of Llama 3.1 | 4–9B | 83.979.6–87.8 | 11.2ROCm 7.1 | 5.9 | 16,384 |
| 47–111 | Qwen3-4B-Thinking-2507 | 4–9B | 83.580.2–87.0 | 21.5ROCm 7.1 | 3.0 | 16,384 |
| 52–118 | Olmo-3-7B-Instruct | 4–9B | 82.978.8–86.9 | 10.2ROCm 7.1 | 5.4 | 16,384 |
| 58–123 | gemma-3-4b-it | 4–9B | 82.177.6–86.5 | 16.9ROCm 7.1 | 4.0 | 16,384 |
| 59–122 | EXAONE-3.5-7.8B-Instruct | 4–9B | 82.077.6–86.4 | 9.5ROCm 7.1 | 5.7 | 16,384 |
| 59–121 | Qwen2.5-VL-7B-Instruct | 4–9B | 82.077.8–86.3 | 10.2ROCm 7.1 | 7.2 | 16,384 |
| 53–118 | Command R7B Arabic 7B 02 2025 | 4–9B | 81.977.5–86.6 | 11.0ROCm 7.1 | 6.1 | 16,384 |
| 59–128 | Granite4.2 8B | 4–9B | 81.977.2–86.3 | 11.1ROCm 7.1 | 6.4 | 16,384 |
| 56–122 | Deepseek R1 8B Llama Distill | 4–9B | 81.776.9–86.1 | 11.9ROCm 7.1 | 5.9 | 16,384 |
| 62–123 | Phi-4-mini-instruct | 3–4B | 81.677.1–86.0 | 17.1ROCm 7.1 | 3.0 | 16,384 |
| 57–133 | Falcon3 10B Instruct | 9–16B | 81.376.6–85.8 | 9.0ROCm 7.1 | 7.6 | 16,384 |
| 65–125 | Llama-3.1-8B-Instruct | 4–9B | 81.076.5–85.2 | 9.6ROCm 7.1 | 5.9 | 16,384 |
| 61–124 | Ministral 3 3B Instruct 2512 | 3–4B | 80.976.5–85.2 | 70.9ROCm 7.1 | 3.6 | 16,384 |
| 66–129 | granite-4.1-3b | 3–4B | 80.775.9–85.3 | 20.1ROCm 7.1 | 2.5 | 16,384 |
| 69–131 | gemma-2-9b-it | 9–16B | 79.975.5–84.2 | 7.6ROCm 7.1 | 6.9 | 8,192 |
| 64–127 | Gemma3n E2B IT | 4–9B | 79.975.2–84.7 | 13.8ROCm 7.1 | 6.7 | 16,384 |
| 67–135 | DeepSeek-R1-0528-Qwen3-8B | 4–9B | 79.675.5–83.9 | 12.2ROCm 7.1 | 6.3 | 16,384 |
| 66–129 | Granite3.3 8B | 4–9B | 79.474.7–83.9 | 11.3ROCm 7.1 | 5.9 | 16,384 |
| 64–132 | Granite4 Tiny H | 4–9B | 79.174.1–84.2 | 24.2ROCm 7.1 | 5.0 | 16,384 |
| 69–136 | DeepSeek-Coder-V2-Lite-Instructcode model | 9–16B | 79.074.3–83.5 | 11.1ROCm 7.1 | 12.4 | 16,384 |
| 72–140 | Granite4.2 3B | 3–4B | 78.273.7–83.1 | 24.3ROCm 7.1 | 2.7 | 16,384 |
| 67–136 | Phi3 14B Medium 128K Instruct | 9–16B | 78.073.2–82.6 | —ROCm 7.1 | 10.3 | 16,384 |
| 68–146 | Olmo-3-7B-Think | 4–9B | 77.773.1–82.5 | 13.5ROCm 7.1 | 5.4 | 16,384 |
| 68–137 | Qwen3 VL 2B Instruct | Tiny (under 3B) | 77.472.5–82.1 | 26.5ROCm 7.1 | 2.3 | 16,384 |
| 76–148 | Qwen2.5-3B-Instruct | 3–4B | 76.872.0–81.6 | 21.9ROCm 7.1 | 2.3 | 16,384 |
| 70–140 | Openthinker 7B V2fine-tune of Qwen 2.5 | 4–9B | 76.872.1–81.6 | 12.6ROCm 7.1 | 5.6 | 16,384 |
| 72–139 | Granite4 Micro H | 3–4B | 76.672.0–81.4 | 13.4ROCm 7.1 | 2.3 | 16,384 |
| 73–141 | Exaone3.5 2.4B Instruct | Tiny (under 3B) | 76.370.9–81.3 | 100.9ROCm 7.1 | 1.9 | 16,384 |
| 71–150 | Falcon3 7B Instruct | 4–9B | 76.271.6–81.3 | 12.5ROCm 7.1 | 5.5 | 16,384 |
| 77–141 | Qwen3 1.7B | Tiny (under 3B) | 76.071.3–80.9 | 38.5ROCm 7.1 | 1.7 | 16,384 |
| 78–142 | Mistral Nemo 12B Instruct 2407 | 9–16B | 75.970.6–80.8 | 8.2ROCm 7.1 | 9.0 | 16,384 |
| 84–150 | Llama-3.2-3B-Instruct | 3–4B | 75.771.0–80.4 | 20.7ROCm 7.1 | 2.4 | 16,384 |
| 80–143 | Lfm2.5 8B A1B | 4–9B | 75.770.8–80.5 | 43.4ROCm 7.1 | 6.2 | 16,384 |
| 81–144 | Qwen3.5 2B | Tiny (under 3B) | 75.169.7–80.3 | 33.3ROCm 7.1 | 2.3 | 16,384 |
| 87–152 | Command R7B (12-2024) | 4–9B | 74.870.1–79.1 | 8.7ROCm 7.1 | 6.1 | 8,192 |
| 87–152 | Meta-Llama-3-8B-Instruct | 4–9B | 74.769.9–79.5 | 9.6ROCm 7.1 | 5.9 | 8,192 |
| 84–146 | Deepcoder 14B Previewfine-tune of DeepSeek-R1-Distill-Qwen 14Bcode model | 9–16B | 74.569.5–79.0 | 6.7ROCm 7.1 | 10.8 | 16,384 |
| 88–152 | medgemma-4b-itmedical model | 4–9B | 74.369.3–79.2 | 17.1ROCm 7.1 | 4.0 | 16,384 |
| 84–148 | Lfm2.5 Thinking 1.2B | Tiny (under 3B) | 74.369.5–79.1 | 205.0ROCm 7.1 | 0.9 | 16,384 |
| 85–145 | Granite4 3B | 3–4B | 74.169.0–79.2 | 71.3ROCm 7.1 | 2.5 | 16,384 |
| 85–145 | Granite3.2 8B Instruct | 4–9B | 74.169.5–78.9 | 34.7ROCm 7.1 | 5.9 | 16,384 |
| 82–155 | DeepSeek-R1-Distill-Qwen-7B | 4–9B | 73.668.9–78.3 | 13.8ROCm 7.1 | 5.6 | 16,384 |
| 87–153 | Glm4 9B Chat | 4–9B | 73.368.0–78.5 | 9.8ROCm 7.1 | 7.6 | 16,384 |
| 87–152 | Tulu3 8Bfine-tune of Llama 3.1 | 4–9B | 73.067.9–77.9 | 12.0ROCm 7.1 | 5.9 | 16,384 |
| 88–151 | Granite3.1 Dense 8B Instruct | 4–9B | 72.967.8–77.6 | 34.8ROCm 7.1 | 6.0 | 16,384 |
| 90–154 | Aya Expanse 8B | 4–9B | 72.567.5–77.3 | 10.7ROCm 7.1 | 6.1 | 8,192 |
| 89–155 | Qwen2.5-Coder-3B-Instructcode model | 3–4B | 72.567.7–77.5 | 21.3ROCm 7.1 | 2.3 | 16,384 |
| 90–157 | Qwen2.5-VL-3B-Instruct | 3–4B | 72.367.2–77.6 | 21.9ROCm 7.1 | 3.8 | 16,384 |
| 89–158 | Lfm3 VL 2B | Tiny (under 3B) | 72.267.2–77.0 | 41.9ROCm 7.1 | 1.8 | 16,384 |
| 91–159 | Medgemma1.5 4B ITfine-tune of Gemma 3medical model | 3–4B | 71.666.6–76.4 | 20.2ROCm 7.1 | 4.0 | 16,384 |
| 97–157 | Phi4 Mini Reasoning 3.8B | 3–4B | 70.965.9–76.0 | 48.8ROCm 7.1 | 3.8 | 16,384 |
| 103–156 | YI 9B Chat V1.5 | 4–9B | 70.565.4–75.3 | 10.0ROCm 7.1 | 6.4 | 4,096 |
| 97–157 | Marco O1 7Bfine-tune of Qwen 2 7B | 4–9B | 70.565.1–76.0 | 11.8ROCm 7.1 | 5.6 | 16,384 |
| 104–157 | Cogito 3B V1 Preview Llamafine-tune of Llama 3.2 | 3–4B | 70.465.4–75.3 | 73.4ROCm 7.1 | 2.6 | 16,384 |
| 96–162 | Lfm3 1.7B | Tiny (under 3B) | 70.265.3–75.3 | 42.1ROCm 7.1 | 1.3 | 16,384 |
| 97–165 | Olmo2 13B 1124 Instruct | 9–16B | 69.664.5–74.5 | 6.8ROCm 7.1 | 10.1 | 4,096 |
| 109–159 | Codegeex4 9B Allcode model | 4–9B | 68.763.2–73.8 | 10.1ROCm 7.1 | 7.6 | 16,384 |
| 111–159 | Granite3.3 2B | Tiny (under 3B) | 68.663.8–73.8 | 28.6ROCm 7.1 | 1.8 | 16,384 |
| 111–160 | Qwen2.5 1.5B Instruct | Tiny (under 3B) | 68.563.2–73.7 | 45.5ROCm 7.1 | 1.2 | 16,384 |
| 112–160 | Gemma2 2B Instruct | Tiny (under 3B) | 68.262.9–73.3 | 27.6ROCm 7.1 | 2.0 | 8,192 |
| 112–160 | Mathstral 7B V0.1fine-tune of Mistral 7Bmath model | 4–9B | 68.062.4–73.5 | 12.0ROCm 7.1 | 5.3 | 16,384 |
| 109–173 | Hermes-3-Llama-3.1-8B | 4–9B | 67.662.4–73.1 | 9.6ROCm 7.1 | 5.9 | 16,384 |
| 117–161 | Granite3 Dense 8B Instruct | 4–9B | 67.662.8–72.3 | 10.5ROCm 7.1 | 5.9 | 4,096 |
| 116–161 | Internlm2 7B Chat V2.5 | 4–9B | 67.462.1–72.6 | 12.0ROCm 7.1 | 5.6 | 16,384 |
| 110–174 | Phi-3.5-mini-instruct | 3–4B | 67.162.0–72.3 | 17.6ROCm 7.1 | 2.9 | 16,384 |
| 119–162 | Granite4 1B H | Tiny (under 3B) | 66.861.4–72.1 | 24.9ROCm 7.1 | 1.9 | 16,384 |
| 108–178 | Qwen2 Math 7B Instructmath model | 4–9B | 66.060.5–71.3 | 12.2ROCm 7.1 | 5.6 | 4,096 |
| 120–170 | Exaone Deep 7.8B | 4–9B | 65.760.0–71.5 | 12.2ROCm 7.1 | 5.8 | 16,384 |
| 125–167 | Olmo2 7B 1124 Instruct | 4–9B | 65.460.6–70.1 | 11.8ROCm 7.1 | 5.4 | 4,096 |
| 113–181 | Falcon3 3B Instruct | 3–4B | 64.859.4–70.6 | 27.9ROCm 7.1 | 2.4 | 16,384 |
| 132–172 | Granite3.2 2B Instruct | Tiny (under 3B) | 63.858.7–69.0 | 30.1ROCm 7.1 | 1.8 | 16,384 |
| 129–182 | Phi-3-mini-4k-instruct | 3–4B | 63.458.3–68.1 | 17.6ROCm 7.1 | 2.9 | 4,096 |
| 137–176 | Phi3 3.8B Mini 128K Instruct | 3–4B | 62.958.0–67.7 | 63.9ROCm 7.1 | 2.9 | 16,384 |
| 143–184 | Smallthinker 3B Previewfine-tune of Qwen 2.5 3B | 3–4B | 61.155.8–66.6 | 22.1ROCm 7.1 | 2.5 | 16,384 |
| 147–183 | Gemma 7B Instruct V1.1 | 4–9B | 60.855.5–66.1 | 10.2ROCm 7.1 | 6.4 | 8,192 |
| 147–184 | Granite3.1 Dense 2B Instruct | Tiny (under 3B) | 60.655.0–66.1 | 28.8ROCm 7.1 | 1.9 | 16,384 |
| 147–185 | Qwen2.5 Coder 1.5B Instructcode model | Tiny (under 3B) | 60.554.9–66.0 | 45.8ROCm 7.1 | 1.2 | 16,384 |
| 148–186 | Mistral 7B Instruct V0.3 | 4–9B | 59.854.7–64.7 | 11.6ROCm 7.1 | 5.3 | 16,384 |
| 148–186 | Granite3.1 Moe 3B Instruct | 3–4B | 59.854.7–65.0 | 53.1ROCm 7.1 | 2.4 | 16,384 |
| 150–186 | Codegemma 7B Instruct V1.1code model | 4–9B | 59.554.8–64.2 | 10.3ROCm 7.1 | 6.4 | 8,192 |
| 143–191 | Mistral-7B-Instruct-v0.2 | 4–9B | 59.254.3–64.2 | 10.2ROCm 7.1 | 5.2 | 16,384 |
| 150–186 | YI Coder 9B Chatcode model | 4–9B | 59.053.7–64.5 | 10.5ROCm 7.1 | 6.4 | 16,384 |
| 151–186 | Solar 10.7B Instruct V1 | 9–16B | 58.753.7–63.8 | 8.7ROCm 7.1 | 7.8 | 4,096 |
| 152–187 | Granite3 Dense 2B Instruct | Tiny (under 3B) | 58.353.1–63.6 | 29.7ROCm 7.1 | 1.9 | 4,096 |
| 152–188 | Deepseek V2 16B Lite Chat | 9–16B | 58.253.2–63.4 | 12.2ROCm 7.1 | 12.0 | 16,384 |
| 152–187 | YI 6B Chat V1.5 | 4–9B | 58.052.8–63.5 | 43.3ROCm 7.1 | 4.4 | 4,096 |
| 150–187 | Llama3.2 1B Instruct | Tiny (under 3B) | 58.052.3–63.7 | 58.5ROCm 7.1 | 1.0 | 16,384 |
| 152–188 | Sailor2 8B Chatfine-tune of Qwen 2.5 | 4–9B | 57.351.7–62.8 | 10.6ROCm 7.1 | 6.2 | 16,384 |
| 154–188 | Gemma3 1B IT | Tiny (under 3B) | 56.851.8–62.2 | 57.0ROCm 7.1 | 1.0 | 16,384 |
| 155–189 | Aya 8B 23 | 4–9B | 55.951.1–60.7 | 10.9ROCm 7.1 | 6.1 | 8,192 |
| 155–192 | Deepcoder 1.5B Previewfine-tune of DeepSeek-R1-Distill-Qwen 1.5Bcode model | Tiny (under 3B) | 55.850.3–61.0 | 44.9ROCm 7.1 | 1.3 | 16,384 |
| 154–195 | Qwen 14B Chat V1.5 | 9–16B | 54.849.8–59.4 | 6.3ROCm 7.1 | 11.0 | 16,384 |
| 156–196 | Exaone Deep 2.4B | Tiny (under 3B) | 54.748.9–60.5 | 31.2ROCm 7.1 | 1.9 | 16,384 |
| 158–194 | Deepseek Llm 7B Chat | 4–9B | 54.349.3–59.8 | 12.2ROCm 7.1 | 5.0 | 4,096 |
| 161–198 | Qwen3 VL 2B Thinking | Tiny (under 3B) | 53.748.3–59.1 | 27.2ROCm 7.1 | 2.3 | 16,384 |
| 159–202 | Codellama 7B Instructcode model | 4–9B | 53.047.8–57.8 | 13.4ROCm 7.1 | 4.9 | 16,384 |
| 159–203 | Codellama 13B Instructcode model | 9–16B | 52.547.2–57.8 | 6.9ROCm 7.1 | 9.5 | 16,384 |
| 168–198 | Deepscaler 1.5B Previewfine-tune of DeepSeek-R1-Distill-Qwen 1.5Bmath model | Tiny (under 3B) | 52.146.6–57.7 | 45.5ROCm 7.1 | 1.3 | 16,384 |
| 171–198 | Nemotron Mini 4B Instruct | 3–4B | 52.047.4–56.9 | 66.8ROCm 7.1 | 3.2 | 4,096 |
| 159–207 | Internlm2 1.8B Chat V2.5 | Tiny (under 3B) | 51.946.1–57.5 | 45.6ROCm 7.1 | 1.4 | 16,384 |
| 175–203 | Qwen2 1.5B Instruct | Tiny (under 3B) | 49.744.1–55.3 | 43.1ROCm 7.1 | 1.2 | 16,384 |
| 176–204 | Codeqwen 7B Chat V1.5code model | 4–9B | 49.444.2–54.7 | 11.8ROCm 7.1 | 5.6 | 16,384 |
| 177–205 | Deepseek R1 1.5B Qwen Distill | Tiny (under 3B) | 48.943.5–54.7 | 44.9ROCm 7.1 | 1.3 | 16,384 |
| 177–206 | Hermes3 3B Llama3.2fine-tune of Llama 3.2 | 3–4B | 48.542.5–54.3 | 23.4ROCm 7.1 | 2.4 | 16,384 |
| 179–206 | Granite3 Moe 3B Instruct | 3–4B | 47.742.3–53.1 | 147.4ROCm 7.1 | 2.5 | 4,096 |
| 180–206 | Smollm2 1.7B Instruct | Tiny (under 3B) | 47.141.7–52.7 | 41.7ROCm 7.1 | 1.3 | 8,192 |
| 184–208 | Granite4 350M H | Tiny (under 3B) | 45.640.7–50.7 | 55.8ROCm 7.1 | 0.4 | 16,384 |
| 181–211 | Llama2 13B Chat | 9–16B | 45.340.2–50.6 | 7.1ROCm 7.1 | 9.5 | 4,096 |
| 180–212 | Falcon3 1B Instruct | Tiny (under 3B) | 44.339.8–49.3 | 51.4ROCm 7.1 | 1.3 | 8,192 |
| 183–211 | Qwen 7B Chat V1.5 | 4–9B | 44.139.1–48.9 | 12.3ROCm 7.1 | 5.8 | 16,384 |
| 185–209 | Dolphin Phi 2.7B V2.6fine-tune of Phi-2 | Tiny (under 3B) | 44.038.9–49.4 | 25.0ROCm 7.1 | 2.2 | 2,048 |
| 185–210 | Qwen3 0.6B | Tiny (under 3B) | 44.038.8–49.4 | 223.0ROCm 7.1 | 0.6 | 16,384 |
| 182–212 | Llama-2-7b-chat-hf | 4–9B | 43.839.0–48.8 | 11.0ROCm 7.1 | 4.9 | 4,096 |
| 185–210 | Stablelm Zephyr 3Bfine-tune of StableLM 3B | Tiny (under 3B) | 43.738.4–49.3 | 26.5ROCm 7.1 | 2.0 | 4,096 |
| 185–212 | Granite3.1 Moe 1B Instruct | Tiny (under 3B) | 43.637.9–49.1 | 102.6ROCm 7.1 | 1.0 | 16,384 |
| 186–212 | Qwen2.5 0.5B Instruct | Tiny (under 3B) | 43.337.4–49.4 | 99.3ROCm 7.1 | 0.5 | 16,384 |
| 190–213 | Stablelm2 12B Chat | 9–16B | 42.438.1–46.7 | 4.7ROCm 7.1 | 8.9 | 4,096 |
| 188–213 | Gemma 2B Instruct V1.1 | Tiny (under 3B) | 41.936.8–47.5 | 29.3ROCm 7.1 | 1.9 | 8,192 |
| 193–215 | Granite Code 8B Instructcode model | 4–9B | 40.735.3–46.7 | 10.5ROCm 7.1 | 5.9 | 4,096 |
| 195–214 | Phi 2.7B Chat V2 | Tiny (under 3B) | 39.935.5–44.2 | 26.0ROCm 7.1 | 2.2 | 2,048 |
| 197–216 | Granite Code 3B Instructcode model | 3–4B | 38.032.7–43.5 | 71.8ROCm 7.1 | 2.5 | 2,048 |
| 201–216 | Qwen2 0.5B Instruct | Tiny (under 3B) | 36.030.9–41.2 | 93.7ROCm 7.1 | 0.5 | 16,384 |
| 203–217 | Granite3 Moe 1B Instruct | Tiny (under 3B) | 35.531.0–40.5 | 70.0ROCm 7.1 | 1.0 | 4,096 |
| 198–219 | Deepseek Coder 6.7B Instructcode model | 4–9B | 34.729.7–40.0 | 13.6ROCm 7.1 | 4.9 | 16,384 |
| 203–222 | Qwen2 Math 1.5B Instructmath model | Tiny (under 3B) | 33.128.0–38.2 | 45.2ROCm 7.1 | 1.2 | 4,096 |
| 208–218 | Qwen2.5 Coder 0.5B Instructcode model | Tiny (under 3B) | 32.827.5–38.2 | 96.7ROCm 7.1 | 0.5 | 16,384 |
| 209–219 | Opencoder 1.5B Instructcode model | Tiny (under 3B) | 31.927.6–36.4 | 17.4ROCm 7.1 | 1.7 | 4,096 |
| 211–219 | Qwen 4B Chat V1.5 | 3–4B | 30.226.0–34.5 | 67.5ROCm 7.1 | 3.0 | 16,384 |
| 212–223 | Smollm2 360M Instruct | Tiny (under 3B) | 28.123.4–32.9 | 275.3ROCm 7.1 | 0.3 | 8,192 |
| 213–224 | YI Coder 1.5B Chatcode model | Tiny (under 3B) | 26.522.1–31.5 | 49.5ROCm 7.1 | 1.2 | 16,384 |
| 216–225 | Qwen 1.8B Chat V1.5 | Tiny (under 3B) | 24.720.2–29.3 | 127.4ROCm 7.1 | 1.4 | 16,384 |
| 216–225 | Stablelm2 1.6B Chat | Tiny (under 3B) | 24.420.2–29.0 | 47.9ROCm 7.1 | 1.2 | 4,096 |
| 217–225 | Sailor2 1B Chatfine-tune of Qwen 2.5 | Tiny (under 3B) | 23.519.0–28.1 | 57.1ROCm 7.1 | 0.9 | 16,384 |
| 219–225 | Stable Code 3B Instructcode model | Tiny (under 3B) | 22.818.4–27.4 | 26.4ROCm 7.1 | 2.0 | 16,384 |
| 218–225 | Deepseek Coder 1.3B Instructcode model | Tiny (under 3B) | 22.618.3–27.2 | 53.6ROCm 7.1 | 1.0 | 16,384 |
| 220–227 | Smollm2 135M Instruct | Tiny (under 3B) | 20.116.2–24.6 | 179.9ROCm 7.1 | 0.1 | 8,192 |
| 225–227 | Qwen 0.5B Chat V1.5 | Tiny (under 3B) | 16.112.5–20.3 | 99.8ROCm 7.1 | 0.5 | 16,384 |
| 225–227 | Codegemma 2B Code V1.1code model | Tiny (under 3B) | 15.311.7–19.1 | 29.1ROCm 7.1 | 1.9 | 8,192 |
Not measured
These ranked models have no model file size on record, so this page cannot estimate their memory need or say whether they fit.
- Ornith-1.5-9B: Overall 89.0, not measured
- SmolLM3-3B: Overall 78.0, not measured
- dolphin-2.9.1-yi-1.5-34b: Overall 73.0, not measured
- Qwen2-VL-7B-Instruct: Overall 72.2, not measured
- Apertus-8B-Instruct-2509: Overall 63.6, not measured
The full speed and memory table is on the Speed page.