llmrun Composite Score · Science
The best LLMs for science, open and proprietary together, on one open-ended scale where 100 is the average model (scientific knowledge). Switch to open models to see only what you can download and run.
46 benchmarks4 sourcesUpdated 14 Nov 2026Reference scale 2026-Q4How the Composite Score is built
Science score vs VRAM needed
This chart doesn't show proprietary models.
| # | Model | llmrun Composite Score | Coding | Agents | Math | Science | Reasoning | Benchmarks | VRAM |
|---|---|---|---|---|---|---|---|---|---|
| 121 | DeepSeek V3.1 Terminusopen412 GB · 5 benchmarks | Coding 107 | Agents – | Math – | Science 99 | Reasoning 108 | 5 | 412 GB | |
| 122 | Mercury 2.54 benchmarks | Coding 95 | Agents – | Math – | Science 99 | Reasoning – | 4 | – | |
| 123 | DeepSeek R1open412 GB · 12 benchmarks | Coding 100 | Agents – | Math – | Science 99 | Reasoning 95 | 12 | 412 GB | |
| 124 | Qwen3 30B A3B Thinking 2507open19.4 GB · 7 benchmarks | Coding – | Agents – | Math – | Science 99 | Reasoning 94 | 7 | 19.4 GB | |
| 125 | DeepSeek v3 0324open412 GB · 12 benchmarks | Coding 100 | Agents – | Math – | Science 98 | Reasoning 88 | 12 | 412 GB | |
| 126 | Gemini 2.5 Pro Preview (May 06)4 benchmarks | Coding – | Agents – | Math – | Science 98 | Reasoning – | 4 | – | |
| 127 | GPT 4.5 Preview (Feb 27, 2025)10 benchmarks | Coding 96 | Agents – | Math – | Science 98 | Reasoning 96 | 10 | – | |
| 128 | NVIDIA Nemotron 3 Super 120B A12B BF16open74.7 GB · 4 benchmarks | Coding 96 | Agents – | Math – | Science 98 | Reasoning – | 4 | 74.7 GB | |
| 129 | Mistral Large 3 675B Instruct 2512open454 GB · 6 benchmarks | Coding 93 | Agents – | Math – | Science 97 | Reasoning 86 | 6 | 454 GB | |
| 130 | GPT 4.1 Mini (Apr 14, 2025)15 benchmarks | Coding 92 | Agents – | Math 95 | Science 97 | Reasoning 91 | 15 | – | |
| 131 | Qwen3 Coder Nextopen48.3 GB · 3 benchmarks | — | Coding 96 | Agents – | Math – | Science 97 | Reasoning – | 3 | 48.3 GB |
| 132 | GPT 5.1 2025 11.137 benchmarks | Coding – | Agents – | Math – | Science 96 | Reasoning 91 | 7 | – | |
| 133 | GPT 4.1 (Apr 14, 2025)16 benchmarks | Coding 99 | Agents – | Math 92 | Science 96 | Reasoning 92 | 16 | – | |
| 134 | Qwen3.5 122B A10Bopen75.5 GB · 6 benchmarks | Coding – | Agents – | Math – | Science 96 | Reasoning 112 | 6 | 75.5 GB | |
| 135 | Qwen3 32Bopen22.6 GB · 8 benchmarks | Coding 93 | Agents – | Math – | Science 96 | Reasoning 92 | 8 | 22.6 GB | |
| 136 | Llama 4 Maverick 17B 128E Instructopen273 GB · 15 benchmarks | Coding 81 | Agents – | Math – | Science 96 | Reasoning 86 | 15 | 273 GB | |
| 137 | Qwen3 14Bopen11.9 GB · 7 benchmarks | Coding – | Agents – | Math – | Science 94 | Reasoning 88 | 7 | 11.9 GB | |
| 138 | GPT OSS 20Bopen13.1 GB · 10 benchmarks | Coding 102 | Agents – | Math – | Science 93 | Reasoning 91 | 10 | 13.1 GB | |
| 139 | Mistral Medium 25086 benchmarks | Coding 92 | Agents – | Math – | Science 91 | Reasoning 92 | 6 | – | |
| 140 | Mistral Medium 25054 benchmarks | Coding – | Agents – | Math – | Science 91 | Reasoning – | 4 | – | |
| 141 | DeepSeek v3open412 GB · 8 benchmarks | Coding 97 | Agents – | Math – | Science 90 | Reasoning 87 | 8 | 412 GB | |
| 142 | Gemini 2.0 Flash Thinking Exp (Jan 21)6 benchmarks | Coding – | Agents – | Math – | Science 89 | Reasoning – | 6 | – | |
| 143 | Gemini 1.5 Pro 0028 benchmarks | Coding – | Agents – | Math – | Science 89 | Reasoning 83 | 8 | – | |
| 144 | Phi 4open12.5 GB · 5 benchmarks | Coding – | Agents – | Math – | Science 89 | Reasoning – | 5 | 12.5 GB | |
| 145 | Claude 3.5 Sonnet (Oct 22, 2024)10 benchmarks | Coding 99 | Agents 100 | Math – | Science 88 | Reasoning 97 | 10 | – | |
| 146 | Qwen3 8Bopen7.6 GB · 7 benchmarks | Coding – | Agents – | Math – | Science 88 | Reasoning 81 | 7 | 7.6 GB | |
| 147 | Devstral Small 2 24B Instruct 2512open18.1 GB · 3 benchmarks | — | Coding – | Agents – | Math – | Science 87 | Reasoning – | 3 | 18.1 GB |
| 148 | Llama 3.1 405B Instructopen276 GB · 8 benchmarks | Coding – | Agents – | Math – | Science 85 | Reasoning 86 | 8 | 276 GB | |
| 149 | Llama 4 Scout 17B 16E Instructopen77.0 GB · 11 benchmarks | Coding – | Agents – | Math – | Science 84 | Reasoning 81 | 11 | 77.0 GB | |
| 150 | Magistral Small 2509open18.1 GB · 5 benchmarks | Coding – | Agents – | Math – | Science 83 | Reasoning – | 5 | 18.1 GB | |
| 151 | MiMo v2 Flashopen187 GB · 3 benchmarks | — | Coding 100 | Agents – | Math – | Science 83 | Reasoning – | 3 | 187 GB |
| 152 | Mistral Small 3.2 24B Instruct 2506open18.1 GB · 5 benchmarks | Coding – | Agents – | Math – | Science 82 | Reasoning – | 5 | 18.1 GB | |
| 153 | Granite 4.1 30Bopen21.9 GB · 2 benchmarks | — | Coding – | Agents – | Math – | Science 82 | Reasoning – | 2 | 21.9 GB |
| 154 | GPT 4.1 Nano (Apr 14, 2025)13 benchmarks | Coding 75 | Agents – | Math – | Science 82 | Reasoning 72 | 13 | – | |
| 155 | Solar Pro 32 benchmarks | — | Coding – | Agents – | Math – | Science 82 | Reasoning – | 2 | – |
| 156 | GPT 4o (Nov 20, 2024)10 benchmarks | Coding 82 | Agents – | Math – | Science 82 | Reasoning 90 | 10 | – | |
| 157 | Mistral Small 3.1 24B Instruct 2503open18.1 GB · 5 benchmarks | Coding – | Agents – | Math – | Science 81 | Reasoning – | 5 | 18.1 GB | |
| 158 | Llama 3.3 70B Instructopen51.3 GB · 12 benchmarks | Coding 76 | Agents – | Math – | Science 81 | Reasoning 83 | 12 | 51.3 GB | |
| 159 | Gemma 3 27B ITopen21.7 GB · 9 benchmarks | Coding 70 | Agents – | Math – | Science 81 | Reasoning 74 | 9 | 21.7 GB | |
| 160 | Claude 3.5 Haiku (Oct 22, 2024)8 benchmarks | Coding 88 | Agents – | Math – | Science 73 | Reasoning – | 8 | – | |
| 161 | Gemma 3 12B ITopen10.9 GB · 7 benchmarks | Coding – | Agents – | Math – | Science 72 | Reasoning 66 | 7 | 10.9 GB | |
| 162 | Phi 4 Mini Instructopen4.8 GB · 3 benchmarks | — | Coding – | Agents – | Math – | Science 62 | Reasoning – | 3 | 4.8 GB |
| 163 | Granite 4.0 Microopen3.7 GB · 3 benchmarks | — | Coding – | Agents – | Math – | Science 62 | Reasoning – | 3 | 3.7 GB |
| 164 | Granite 4.0 350Mopen1.0 GB · 3 benchmarks | — | Coding – | Agents – | Math – | Science 58 | Reasoning – | 3 | 1.0 GB |
| 165 | Llama 3.1 8B Instructopen7.9 GB · 9 benchmarks | Coding 56 | Agents – | Math – | Science 56 | Reasoning 70 | 9 | 7.9 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.
Dataset. Grab the scores of every open model as a dataset on Hugging Face, including a timestamped snapshot for each refresh.