Best model for reasoning
Models ranked by their reasoning 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 97.1 on a scale of 0 to 100; its Overall is 89.1 (95% interval 86.0–91.9).
| Place | Model | Reasoning (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | DeepSeek-R1-Distill-Qwen-32B | 97.135 of 35 items scored | 89.186.0–91.9 | 19–85 | 3.1ROCm 7.1 | 23.8 | 16–35B |
| 1 | gemma-4-31B-it | 97.135 of 35 items scored | 99.398.3–100.0 | 1–5 | 3.8ROCm 7.1 | 22.4 | 16–35B |
| 1 | gpt-oss-120b | 97.135 of 35 items scored | 98.696.8–100.0 | 1–14 | 11.0Vulkan | 78.4 | > 35B |
| 1 | gpt-oss-20b | 97.135 of 35 items scored | 95.092.2–97.6 | 2–60 | 14.9ROCm 7.1 | 16.6 | 16–35B |
| 1 | Llama-3.3-70B-Instruct | 97.135 of 35 items scored | 94.391.3–96.7 | 4–63 | 1.4Vulkan | 51.0 | > 35B |
Tiny (under 3B)
Exaone Deep 2.4B leads its size class with 82.9 on a scale of 0 to 100; its Overall is 54.7 (95% interval 48.9–60.5).
| Place | Model | Reasoning (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | Exaone Deep 2.4B | 82.935 of 35 items scored | 54.748.9–60.5 | 156–196 | 31.2ROCm 7.1 | 1.9 | Tiny (under 3B) |
| 1 | Exaone3.5 2.4B Instruct | 82.935 of 35 items scored | 76.370.9–81.3 | 73–141 | 100.9ROCm 7.1 | 1.9 | Tiny (under 3B) |
| 1 | Lfm2.5 Thinking 1.2B | 82.935 of 35 items scored | 74.369.5–79.1 | 84–148 | 205.0ROCm 7.1 | 0.9 | Tiny (under 3B) |
| 4 | Deepcoder 1.5B Previewfine-tune of DeepSeek-R1-Distill-Qwen 1.5Bcode model | 80.035 of 35 items scored | 55.850.3–61.0 | 155–192 | 44.9ROCm 7.1 | 1.3 | Tiny (under 3B) |
| 5 | Lfm3 1.7B | 74.335 of 35 items scored | 70.265.3–75.3 | 96–162 | 42.1ROCm 7.1 | 1.3 | Tiny (under 3B) |
3–4B
Phi4 Mini Reasoning 3.8B leads its size class with 94.3 on a scale of 0 to 100; its Overall is 70.9 (95% interval 65.9–76.0).
| Place | Model | Reasoning (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | Phi4 Mini Reasoning 3.8B | 94.335 of 35 items scored | 70.965.9–76.0 | 97–157 | 48.8ROCm 7.1 | 3.8 | 3–4B |
| 2 | Spark-X2.6-4B | 85.735 of 35 items scored | 85.080.9–89.1 | 43–106 | 19.7ROCm 7.1 | 3.0 | 3–4B |
| 3 | Phi-3-mini-4k-instruct | 77.135 of 35 items scored | 63.458.3–68.1 | 129–182 | 17.6ROCm 7.1 | 2.9 | 3–4B |
| 3 | Phi-4-mini-instruct | 77.135 of 35 items scored | 81.677.1–86.0 | 62–123 | 17.1ROCm 7.1 | 3.0 | 3–4B |
| 5 | Granite4.2 3B | 74.335 of 35 items scored | 78.273.7–83.1 | 72–140 | 24.3ROCm 7.1 | 2.7 | 3–4B |
4–9B
Qwen3-8B leads its size class with 94.3 on a scale of 0 to 100; its Overall is 91.5 (95% interval 88.4–94.5).
| Place | Model | Reasoning (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | Qwen3-8B | 94.335 of 35 items scored | 91.588.4–94.5 | 11–72 | 9.3ROCm 7.1 | 6.3 | 4–9B |
| 2 | Deepseek R1 8B Llama Distill | 91.435 of 35 items scored | 81.776.9–86.1 | 56–122 | 11.9ROCm 7.1 | 5.9 | 4–9B |
| 2 | Granite4.1 8B | 91.435 of 35 items scored | 92.189.0–95.0 | 6–73 | 10.4ROCm 7.1 | 6.4 | 4–9B |
| 2 | Ornith 9B | 91.435 of 35 items scored | 90.086.5–93.1 | 12–79 | 10.8ROCm 7.1 | 6.7 | 4–9B |
| 2 | Qwen3-VL-4B-Instruct | 91.435 of 35 items scored | 89.986.4–93.1 | 12–76 | 17.3ROCm 7.1 | 4.0 | 4–9B |
9–16B
Gemma 4 12B IT leads its size class with 94.3 on a scale of 0 to 100; its Overall is 98.6 (95% interval 97.0–99.8).
| Place | Model | Reasoning (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | Gemma 4 12B IT | 94.335 of 35 items scored | 98.697.0–99.8 | 1–12 | 8.1ROCm 7.1 | 8.9 | 9–16B |
| 2 | DeepSeek-R1-Distill-Qwen-14B | 91.435 of 35 items scored | 85.381.2–89.2 | 38–106 | 7.1ROCm 7.1 | 10.8 | 9–16B |
| 2 | Ornith-1.5-9B | 91.435 of 35 items scored | 89.085.4–92.4 | 19–81 | 8.1ROCm 7.1 | — | 9–16B |
| 2 | phi-4 | 91.435 of 35 items scored | 85.981.9–89.8 | 39–99 | 5.0ROCm 7.1 | 10.9 | 9–16B |
| 2 | Qwen3-14B | 91.435 of 35 items scored | 91.788.5–94.7 | 11–71 | 5.0ROCm 7.1 | 11.1 | 9–16B |
16–35B
DeepSeek-R1-Distill-Qwen-32B leads its size class with 97.1 on a scale of 0 to 100; its Overall is 89.1 (95% interval 86.0–91.9).
| Place | Model | Reasoning (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | DeepSeek-R1-Distill-Qwen-32B | 97.135 of 35 items scored | 89.186.0–91.9 | 19–85 | 3.1ROCm 7.1 | 23.8 | 16–35B |
| 1 | gemma-4-31B-it | 97.135 of 35 items scored | 99.398.3–100.0 | 1–5 | 3.8ROCm 7.1 | 22.4 | 16–35B |
| 1 | gpt-oss-20b | 97.135 of 35 items scored | 95.092.2–97.6 | 2–60 | 14.9ROCm 7.1 | 16.6 | 16–35B |
| 4 | gemma-4-26B-A4B-it | 94.335 of 35 items scored | 97.895.8–99.3 | 1–22 | 15.7ROCm 7.1 | 21.6 | 16–35B |
| 4 | Muse Glimmer 30B | 94.335 of 35 items scored | 97.996.0–99.3 | 1–23 | 3.6ROCm 7.1 | 21.8 | 16–35B |
> 35B
gpt-oss-120b leads its size class with 97.1 on a scale of 0 to 100; its Overall is 98.6 (95% interval 96.8–100.0).
| Place | Model | Reasoning (0 to 100) | Overall | Rank range | Decode (tok/s) | Memory needed (GB, est.) | Size class |
|---|---|---|---|---|---|---|---|
| 1 | gpt-oss-120b | 97.135 of 35 items scored | 98.696.8–100.0 | 1–14 | 11.0Vulkan | 78.4 | > 35B |
| 1 | Llama-3.3-70B-Instruct | 97.135 of 35 items scored | 94.391.3–96.7 | 4–63 | 1.4Vulkan | 51.0 | > 35B |
| 3 | Llama 4 Scout 17B-16E Instruct | 91.435 of 35 items scored | 93.690.7–96.1 | 4–65 | 17.0Vulkan | 80.9 | > 35B |
| 4 | Qwen3-Coder-Nextcode model | 88.635 of 35 items scored | 96.494.2–98.3 | 2–42 | 18.1Vulkan | 62.1 | > 35B |
| 4 | Sage-A1.5-Q4_K_M | 88.635 of 35 items scored | 94.591.8–96.7 | 5–56 | 22.6ROCm 7.1 | 26.4 | > 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.