Best model for long context
Models ranked by their long context score, 0 to 100, as on the category page. Only ranked models with enough scored items are included, and models are grouped by total parameters.
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; 42 of the models ranked ran at a shorter context. Speeds were measured on one laptop and are comparable only between models on the same backend. See the settings and the memory method.
All sizes
DeepSeek-R1-Distill-Qwen-32B leads here with 100.0 on a scale of 0 to 100; its Overall is 89.1 (95% interval 86.0–91.9).
| Place | Model | Long context (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | DeepSeek-R1-Distill-Qwen-32B | 100.018 of 18 items scored | 89.186.0–91.9 | 19–85 | 3.1ROCm 7.1 | 23.8 | 16–35B |
| 1 | gemma-3-27b-it | 100.018 of 18 items scored | 95.893.7–97.5 | 3–36 | 2.7ROCm 7.1 | 20.9 | 16–35B |
| 1 | gemma-4-31B-it | 100.018 of 18 items scored | 99.398.3–100.0 | 1–5 | 3.8ROCm 7.1 | 22.4 | 16–35B |
| 1 | Gemma4 E2B IT | 100.018 of 18 items scored | 93.891.0–96.1 | 4–61 | 17.2ROCm 7.1 | 5.5 | 4–9B |
| 1 | Ministral 3 14B Instruct (2512) | 100.018 of 18 items scored | 92.790.0–95.2 | 9–61 | 5.6ROCm 7.1 | 10.9 | 9–16B |
Tiny (under 3B)
Qwen3.5 2B leads its size class with 77.8 on a scale of 0 to 100; its Overall is 75.1 (95% interval 69.7–80.3).
| Place | Model | Long context (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | Qwen3.5 2B | 77.818 of 18 items scored | 75.169.7–80.3 | 81–144 | 33.3ROCm 7.1 | 2.3 | Tiny (under 3B) |
| 2 | Qwen2.5 1.5B Instruct | 66.718 of 18 items scored | 68.563.2–73.7 | 111–160 | 45.5ROCm 7.1 | 1.2 | Tiny (under 3B) |
| 2 | Qwen2.5 Coder 1.5B Instructcode model | 66.718 of 18 items scored | 60.554.9–66.0 | 147–185 | 45.8ROCm 7.1 | 1.2 | Tiny (under 3B) |
| 2 | Qwen3 VL 2B Instruct | 66.718 of 18 items scored | 77.472.5–82.1 | 68–137 | 26.5ROCm 7.1 | 2.3 | Tiny (under 3B) |
| 5 | Gemma2 2B Instruct | 55.618 of 18 items scored | 68.262.9–73.3 | 112–160 | 27.6ROCm 7.1 | 2.0 | Tiny (under 3B) |
3–4B
SmolLM3-3B leads its size class with 77.8 on a scale of 0 to 100; its Overall is 78.0 (95% interval 73.0–83.1).
| Place | Model | Long context (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | SmolLM3-3B | 77.818 of 18 items scored | 78.073.0–83.1 | 71–142 | 22.2ROCm 7.1 | — | 3–4B |
| 1 | Spark-X2.6-4B | 77.818 of 18 items scored | 85.080.9–89.1 | 43–106 | 19.7ROCm 7.1 | 3.0 | 3–4B |
| 3 | Medgemma1.5 4B ITfine-tune of Gemma 3medical model | 72.218 of 18 items scored | 71.666.6–76.4 | 91–159 | 20.2ROCm 7.1 | 4.0 | 3–4B |
| 3 | Phi-3.5-mini-instruct | 72.218 of 18 items scored | 67.162.0–72.3 | 110–174 | 17.6ROCm 7.1 | 2.9 | 3–4B |
| 3 | Qwen2.5-Coder-3B-Instructcode model | 72.218 of 18 items scored | 72.567.7–77.5 | 89–155 | 21.3ROCm 7.1 | 2.3 | 3–4B |
4–9B
Gemma4 E2B IT leads its size class with 100.0 on a scale of 0 to 100; its Overall is 93.8 (95% interval 91.0–96.1).
| Place | Model | Long context (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | Gemma4 E2B IT | 100.018 of 18 items scored | 93.891.0–96.1 | 4–61 | 17.2ROCm 7.1 | 5.5 | 4–9B |
| 2 | Gemma4 E4B IT | 94.418 of 18 items scored | 95.593.0–97.7 | 2–51 | 13.2ROCm 7.1 | 7.9 | 4–9B |
| 2 | Granite4.1 8B | 94.418 of 18 items scored | 92.189.0–95.0 | 6–73 | 10.4ROCm 7.1 | 6.4 | 4–9B |
| 2 | Ministral 3 8B Instruct 2512 | 94.418 of 18 items scored | 88.785.2–92.1 | 19–86 | 11.0ROCm 7.1 | 7.2 | 4–9B |
| 2 | Ornith 9B | 94.418 of 18 items scored | 90.086.5–93.1 | 12–79 | 10.8ROCm 7.1 | 6.7 | 4–9B |
9–16B
Ministral 3 14B Instruct (2512) leads its size class with 100.0 on a scale of 0 to 100; its Overall is 92.7 (95% interval 90.0–95.2).
| Place | Model | Long context (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | Ministral 3 14B Instruct (2512) | 100.018 of 18 items scored | 92.790.0–95.2 | 9–61 | 5.6ROCm 7.1 | 10.9 | 9–16B |
| 2 | Gemma 4 12B IT | 94.418 of 18 items scored | 98.697.0–99.8 | 1–12 | 8.1ROCm 7.1 | 8.9 | 9–16B |
| 2 | Qwen2.5-14B-Instruct | 94.418 of 18 items scored | 91.388.0–94.2 | 11–72 | 21.7ROCm 7.1 | 10.8 | 9–16B |
| 4 | Cogito 14B V1 Preview Qwenfine-tune of Qwen 2.5 | 88.918 of 18 items scored | 90.386.8–93.4 | 11–79 | 6.6ROCm 7.1 | 10.8 | 9–16B |
| 4 | Ornith-1.5-9B | 88.918 of 18 items scored | 89.085.4–92.4 | 19–81 | 8.1ROCm 7.1 | — | 9–16B |
16–35B
DeepSeek-R1-Distill-Qwen-32B leads its size class with 100.0 on a scale of 0 to 100; its Overall is 89.1 (95% interval 86.0–91.9).
| Place | Model | Long context (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | DeepSeek-R1-Distill-Qwen-32B | 100.018 of 18 items scored | 89.186.0–91.9 | 19–85 | 3.1ROCm 7.1 | 23.8 | 16–35B |
| 1 | gemma-3-27b-it | 100.018 of 18 items scored | 95.893.7–97.5 | 3–36 | 2.7ROCm 7.1 | 20.9 | 16–35B |
| 1 | gemma-4-31B-it | 100.018 of 18 items scored | 99.398.3–100.0 | 1–5 | 3.8ROCm 7.1 | 22.4 | 16–35B |
| 1 | Qwen3.5-27B | 100.018 of 18 items scored | 96.293.8–98.3 | 3–35 | 2.7ROCm 7.1 | 19.9 | 16–35B |
| 1 | Qwen3.5-35B-A3B | 100.018 of 18 items scored | 94.892.1–97.1 | 3–60 | 23.2Vulkan | 25.4 | 16–35B |
> 35B
gpt-oss-120b leads its size class with 94.4 on a scale of 0 to 100; its Overall is 98.6 (95% interval 96.8–100.0).
| Place | Model | Long context (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | gpt-oss-120b | 94.418 of 18 items scored | 98.696.8–100.0 | 1–14 | 11.0Vulkan | 78.4 | > 35B |
| 1 | Llama 4 Scout 17B-16E Instruct | 94.418 of 18 items scored | 93.690.7–96.1 | 4–65 | 17.0Vulkan | 80.9 | > 35B |
| 1 | Qwen3-Coder-Nextcode model | 94.418 of 18 items scored | 96.494.2–98.3 | 2–42 | 18.1Vulkan | 62.1 | > 35B |
| 1 | Sage-A1.5-Q4_K_M | 94.418 of 18 items scored | 94.591.8–96.7 | 5–56 | 22.6ROCm 7.1 | 26.4 | > 35B |
| 5 | Llama-3.3-70B-Instruct | 88.918 of 18 items scored | 94.391.3–96.7 | 4–63 | 1.4Vulkan | 51.0 | > 35B |
A higher place is not always a clear win. When two models have overlapping rank ranges, this battery cannot tell them apart; see how sure a score is. A dash means not measured or not scored, which is different from 0. Speeds were measured on one laptop and are comparable only between models on the same backend.
More: all the best-model pages, the hardware picker, speed and memory for every model.