llmrun Composite Score · Reasoning
The best open LLMs for reasoning, ranked on hard multi-step reasoning boards such as ARC-AGI and HLE. The score shown is the overall llmrun Composite Score; the highlighted column is the reasoning skill score used for the order.
45 benchmarks4 sourcesUpdated 17 Oct 2026Reference scale 2026-Q4How the Composite Score is built
Reasoning score vs VRAM needed
This chart doesn't show proprietary models.
| # (open / overall) | Model | llmrun Composite Score | Coding | Agents | Math | Science | Reasoning | Benchmarks | VRAM |
|---|---|---|---|---|---|---|---|---|---|
| 1 / 58 | Qwen3.8 27B17.9 GB · 15 benchmarks | 120 | 123 | 118 | 112 | 124 | 15 | 17.9 GB | |
| 2 / 103 | Qwen3.5 27B17.9 GB · 6 benchmarks | 98 | – | – | – | 111 | 6 | 17.9 GB | |
| 3 / 104 | Gemma 4 31B IT21.7 GB · 13 benchmarks | 108 | – | – | 105 | 110 | 13 | 21.7 GB | |
| 4 / 116 | Qwen3.5 35B A3B22.0 GB · 7 benchmarks | – | – | – | 108 | 107 | 7 | 22.0 GB | |
| 5 / 123 | Qwen3.6 27B17.9 GB · 11 benchmarks | 102 | – | 114 | 111 | 105 | 11 | 17.9 GB | |
| 6 / 125 | Qwen3.6 35B A3B22.0 GB · 11 benchmarks | 105 | – | 109 | 110 | 104 | 11 | 22.0 GB | |
| 7 / 130 | Gemma 4 26B A4B IT16.5 GB · 8 benchmarks | 104 | – | – | 102 | 102 | 8 | 16.5 GB | |
| 8 / 139 | Qwen3.5 9B6.6 GB · 8 benchmarks | – | – | – | 105 | 97 | 8 | 6.6 GB | |
| 9 / 147 | Qwen3 30B A3B Thinking 250719.4 GB · 7 benchmarks | – | – | – | 99 | 94 | 7 | 19.4 GB | |
| 10 / 149 | Qwen3 30B A3B Instruct 250719.4 GB · 4 benchmarks | – | – | – | – | 93 | 4 | 19.4 GB | |
| 11 / 154 | Qwen3 32B22.6 GB · 8 benchmarks | 93 | – | – | 96 | 92 | 8 | 22.6 GB | |
| 12 / 155 | GPT OSS 20B13.1 GB · 10 benchmarks | 101 | – | – | 92 | 91 | 10 | 13.1 GB | |
| 13 / 162 | Qwen3 14B11.9 GB · 7 benchmarks | – | – | – | 94 | 88 | 7 | 11.9 GB | |
| 14 / 163 | Magistral Small 250617.8 GB · 4 benchmarks | – | – | – | – | 88 | 4 | 17.8 GB | |
| 15 / 173 | Qwen2.5 72B Instruct49.3 GB · 8 benchmarks | – | – | – | – | 86 | 8 | 49.3 GB | |
| 16 / 179 | Qwen3 30B A3B19.4 GB · 6 benchmarks | – | – | – | – | 84 | 6 | 19.4 GB | |
| 17 / 182 | Llama 3.1 70B Instruct51.3 GB · 7 benchmarks | – | – | – | – | 83 | 7 | 51.3 GB | |
| 18 / 185 | Llama 3.3 70B Instruct51.3 GB · 12 benchmarks | 76 | – | – | 81 | 82 | 12 | 51.3 GB | |
| 19 / 186 | Qwen3 8B7.6 GB · 7 benchmarks | – | – | – | 88 | 81 | 7 | 7.6 GB | |
| 20 / 192 | Gemma 3 27B IT21.7 GB · 9 benchmarks | 70 | – | – | 81 | 74 | 9 | 21.7 GB | |
| 21 / 197 | Llama 3.1 8B Instruct7.9 GB · 9 benchmarks | 56 | – | – | 56 | 69 | 9 | 7.9 GB | |
| 22 / 198 | Mixtral 8x7B Instruct v0.130.5 GB · 4 benchmarks | – | – | – | – | 69 | 4 | 30.5 GB | |
| 23 / 199 | Gemma 3 4B IT5.0 GB · 5 benchmarks | – | – | – | – | 69 | 5 | 5.0 GB | |
| 24 / 202 | Gemma 3 12B IT10.9 GB · 7 benchmarks | – | – | – | 71 | 66 | 7 | 10.9 GB | |
| 25 / 204 | Qwen2.5 7B Instruct5.8 GB · 5 benchmarks | – | – | – | – | 65 | 5 | 5.8 GB | |
| 26 / 206 | C4ai Command R 08 202425.1 GB · 2 benchmarks | — | – | – | – | – | 64 | 2 | 25.1 GB |
| 27 / 207 | Meta Llama 3 8B Instruct7.9 GB · 5 benchmarks | – | – | – | – | 64 | 5 | 7.9 GB | |
| 28 / 209 | Llama 2 70B Chat HF50.2 GB · 4 benchmarks | – | – | – | – | 45 | 4 | 50.2 GB |
The range after each score is a 90% interval: given the boards a model has results on, its true score is very likely inside it. Short ranges mean many agreeing results; long ranges mean few or conflicting ones.
Why some models are missing. A model receives a score only if it has results on at least 4 benchmarks across at least 2 categories, with one being reasoning or coding, or on 3 benchmarks that narrow its 90% range to at most 15.6 points. The results must come from at least two different benchmarks: categories of one benchmark, such as LiveBench, count once. Skill scores require at least 2 benchmarks for that skill; otherwise they display as –.
VRAM is llmrun's estimate at Q4_K_M (or the smallest quantization we track) for the weights plus room for a 16K-token context. Longer contexts need more memory. Proprietary models can't run locally, so the VRAM filter hides them.
How the llmrun Composite Score is built · llmrun doesn't run these benchmarks; scores come from public sources.