llmrun Composite Score
The best open LLMs: every open model we can rank, on one open-ended quality scale fitted across public leaderboards, where 100 is the average model. Filter to the models that fit your GPU.
45 benchmarks4 sourcesUpdated 17 Oct 2026Reference scale 2026-Q4How the Composite Score is built
Composite 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 / 81 | Qwen3.8 27B17.9 GB · 15 benchmarks | 120 | 123 | 118 | 112 | 124 | 15 | 17.9 GB | |
| 2 / 112 | Qwen3.6 27B17.9 GB · 11 benchmarks | 102 | – | 114 | 111 | 105 | 11 | 17.9 GB | |
| 3 / 137 | Gemma 4 31B IT21.7 GB · 13 benchmarks | 108 | – | – | 105 | 110 | 13 | 21.7 GB | |
| 4 / 138 | Qwen3.6 35B A3B22.0 GB · 11 benchmarks | 105 | – | 109 | 110 | 104 | 11 | 22.0 GB | |
| 5 / 143 | Qwen3.5 35B A3B22.0 GB · 7 benchmarks | – | – | – | 108 | 107 | 7 | 22.0 GB | |
| 6 / 144 | Gemma 4 26B A4B IT16.5 GB · 8 benchmarks | 104 | – | – | 102 | 102 | 8 | 16.5 GB | |
| 7 / 156 | Qwen3.5 27B17.9 GB · 6 benchmarks | 98 | – | – | – | 111 | 6 | 17.9 GB | |
| 8 / 161 | Seed OSS 36B Instruct24.1 GB · 3 benchmarks | – | – | – | – | – | 3 | 24.1 GB | |
| 9 / 166 | Qwen3 30B A3B Thinking 250719.4 GB · 7 benchmarks | – | – | – | 99 | 94 | 7 | 19.4 GB | |
| 10 / 173 | Qwen3.5 9B6.6 GB · 8 benchmarks | – | – | – | 105 | 97 | 8 | 6.6 GB | |
| 11 / 176 | Qwen3 32B22.6 GB · 8 benchmarks | 93 | – | – | 96 | 92 | 8 | 22.6 GB | |
| 12 / 180 | Qwen3 14B11.9 GB · 7 benchmarks | – | – | – | 94 | 88 | 7 | 11.9 GB | |
| 13 / 182 | DeepSeek R1 Distill Qwen 32B24.3 GB · 3 benchmarks | – | – | – | – | – | 3 | 24.3 GB | |
| 14 / 183 | Qwen3 30B A3B Instruct 250719.4 GB · 4 benchmarks | – | – | – | – | 93 | 4 | 19.4 GB | |
| 15 / 186 | GPT OSS 20B13.1 GB · 10 benchmarks | 101 | – | – | 92 | 91 | 10 | 13.1 GB | |
| 16 / 187 | QwQ 32B24.3 GB · 5 benchmarks | – | – | – | – | – | 5 | 24.3 GB | |
| 17 / 192 | Qwen3 8B7.6 GB · 7 benchmarks | – | – | – | 88 | 81 | 7 | 7.6 GB | |
| 18 / 193 | Qwen3 30B A3B19.4 GB · 6 benchmarks | – | – | – | – | 84 | 6 | 19.4 GB | |
| 19 / 209 | Magistral Small 250617.8 GB · 4 benchmarks | – | – | – | – | 88 | 4 | 17.8 GB | |
| 20 / 213 | Mistral Small 3.2 24B Instruct 250618.1 GB · 5 benchmarks | – | – | – | 82 | – | 5 | 18.1 GB | |
| 21 / 214 | Magistral Small 250918.1 GB · 5 benchmarks | – | – | – | 83 | – | 5 | 18.1 GB | |
| 22 / 217 | Phi 412.5 GB · 4 benchmarks | – | – | – | – | – | 4 | 12.5 GB | |
| 23 / 221 | Qwen2.5 Coder 32B Instruct24.3 GB · 3 benchmarks | – | 84 | – | – | – | 3 | 24.3 GB | |
| 24 / 224 | Qwen2.5 72B Instruct49.3 GB · 8 benchmarks | – | – | – | – | 86 | 8 | 49.3 GB | |
| 25 / 227 | Qwen2.5 32B Instruct24.3 GB · 3 benchmarks | – | – | – | – | – | 3 | 24.3 GB | |
| 26 / 231 | Llama 3.3 70B Instruct51.3 GB · 12 benchmarks | 76 | – | – | 81 | 82 | 12 | 51.3 GB | |
| 27 / 233 | Gemma 3 27B IT21.7 GB · 9 benchmarks | 70 | – | – | 81 | 74 | 9 | 21.7 GB | |
| 28 / 234 | Mistral Small 3.1 24B Instruct 250318.1 GB · 5 benchmarks | – | – | – | 81 | – | 5 | 18.1 GB | |
| 29 / 239 | Llama 3.1 70B Instruct51.3 GB · 7 benchmarks | – | – | – | – | 83 | 7 | 51.3 GB | |
| 30 / 240 | Qwen2 72B Instruct49.3 GB · 3 benchmarks | – | – | – | – | – | 3 | 49.3 GB | |
| 31 / 241 | Meta Llama 3 70B Instruct51.3 GB · 4 benchmarks | – | – | – | – | – | 4 | 51.3 GB | |
| 32 / 242 | Gemma 3 12B IT10.9 GB · 7 benchmarks | – | – | – | 71 | 66 | 7 | 10.9 GB | |
| 33 / 245 | Gemma 2 27B IT19.5 GB · 3 benchmarks | – | – | – | – | – | 3 | 19.5 GB | |
| 34 / 250 | Gemma 2 9B IT7.3 GB · 3 benchmarks | – | – | – | – | – | 3 | 7.3 GB | |
| 35 / 252 | Mistral Nemo Instruct 240711.0 GB · 4 benchmarks | – | – | – | – | – | 4 | 11.0 GB | |
| 36 / 253 | Qwen2.5 7B Instruct5.8 GB · 5 benchmarks | – | – | – | – | 65 | 5 | 5.8 GB | |
| 37 / 255 | Mixtral 8x7B Instruct v0.130.5 GB · 4 benchmarks | – | – | – | – | 69 | 4 | 30.5 GB | |
| 38 / 256 | Gemma 3 4B IT5.0 GB · 5 benchmarks | – | – | – | – | 69 | 5 | 5.0 GB | |
| 39 / 257 | Llama 3.1 8B Instruct7.9 GB · 9 benchmarks | 56 | – | – | 56 | 69 | 9 | 7.9 GB | |
| 40 / 258 | Meta Llama 3 8B Instruct7.9 GB · 5 benchmarks | – | – | – | – | 64 | 5 | 7.9 GB | |
| 41 / 260 | 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.