llmrun

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.

Models ranked by science llmrun Composite Score
#Modelllmrun Composite ScoreCodingAgentsMathScienceReasoningBenchmarksVRAM
121DeepSeek V3.1 Terminusopen412 GB · 5 benchmarksCoding 107Agents –Math –Science 99Reasoning 1085412 GB
122Mercury 2.54 benchmarksCoding 95Agents –Math –Science 99Reasoning –4–
123DeepSeek R1open412 GB · 12 benchmarksCoding 100Agents –Math –Science 99Reasoning 9512412 GB
124Qwen3 30B A3B Thinking 2507open19.4 GB · 7 benchmarksCoding –Agents –Math –Science 99Reasoning 94719.4 GB
125DeepSeek v3 0324open412 GB · 12 benchmarksCoding 100Agents –Math –Science 98Reasoning 8812412 GB
126Gemini 2.5 Pro Preview (May 06)4 benchmarksCoding –Agents –Math –Science 98Reasoning –4–
127GPT 4.5 Preview (Feb 27, 2025)10 benchmarksCoding 96Agents –Math –Science 98Reasoning 9610–
128NVIDIA Nemotron 3 Super 120B A12B BF16open74.7 GB · 4 benchmarksCoding 96Agents –Math –Science 98Reasoning –474.7 GB
129Mistral Large 3 675B Instruct 2512open454 GB · 6 benchmarksCoding 93Agents –Math –Science 97Reasoning 866454 GB
130GPT 4.1 Mini (Apr 14, 2025)15 benchmarksCoding 92Agents –Math 95Science 97Reasoning 9115–
131Qwen3 Coder Nextopen48.3 GB · 3 benchmarks—Coding 96Agents –Math –Science 97Reasoning –348.3 GB
132GPT 5.1 2025 11.137 benchmarksCoding –Agents –Math –Science 96Reasoning 917–
133GPT 4.1 (Apr 14, 2025)16 benchmarksCoding 99Agents –Math 92Science 96Reasoning 9216–
134Qwen3.5 122B A10Bopen75.5 GB · 6 benchmarksCoding –Agents –Math –Science 96Reasoning 112675.5 GB
135Qwen3 32Bopen22.6 GB · 8 benchmarksCoding 93Agents –Math –Science 96Reasoning 92822.6 GB
136Llama 4 Maverick 17B 128E Instructopen273 GB · 15 benchmarksCoding 81Agents –Math –Science 96Reasoning 8615273 GB
137Qwen3 14Bopen11.9 GB · 7 benchmarksCoding –Agents –Math –Science 94Reasoning 88711.9 GB
138GPT OSS 20Bopen13.1 GB · 10 benchmarksCoding 102Agents –Math –Science 93Reasoning 911013.1 GB
139Mistral Medium 25086 benchmarksCoding 92Agents –Math –Science 91Reasoning 926–
140Mistral Medium 25054 benchmarksCoding –Agents –Math –Science 91Reasoning –4–
141DeepSeek v3open412 GB · 8 benchmarksCoding 97Agents –Math –Science 90Reasoning 878412 GB
142Gemini 2.0 Flash Thinking Exp (Jan 21)6 benchmarksCoding –Agents –Math –Science 89Reasoning –6–
143Gemini 1.5 Pro 0028 benchmarksCoding –Agents –Math –Science 89Reasoning 838–
144Phi 4open12.5 GB · 5 benchmarksCoding –Agents –Math –Science 89Reasoning –512.5 GB
145Claude 3.5 Sonnet (Oct 22, 2024)10 benchmarksCoding 99Agents 100Math –Science 88Reasoning 9710–
146Qwen3 8Bopen7.6 GB · 7 benchmarksCoding –Agents –Math –Science 88Reasoning 8177.6 GB
147Devstral Small 2 24B Instruct 2512open18.1 GB · 3 benchmarks—Coding –Agents –Math –Science 87Reasoning –318.1 GB
148Llama 3.1 405B Instructopen276 GB · 8 benchmarksCoding –Agents –Math –Science 85Reasoning 868276 GB
149Llama 4 Scout 17B 16E Instructopen77.0 GB · 11 benchmarksCoding –Agents –Math –Science 84Reasoning 811177.0 GB
150Magistral Small 2509open18.1 GB · 5 benchmarksCoding –Agents –Math –Science 83Reasoning –518.1 GB
151MiMo v2 Flashopen187 GB · 3 benchmarks—Coding 100Agents –Math –Science 83Reasoning –3187 GB
152Mistral Small 3.2 24B Instruct 2506open18.1 GB · 5 benchmarksCoding –Agents –Math –Science 82Reasoning –518.1 GB
153Granite 4.1 30Bopen21.9 GB · 2 benchmarks—Coding –Agents –Math –Science 82Reasoning –221.9 GB
154GPT 4.1 Nano (Apr 14, 2025)13 benchmarksCoding 75Agents –Math –Science 82Reasoning 7213–
155Solar Pro 32 benchmarks—Coding –Agents –Math –Science 82Reasoning –2–
156GPT 4o (Nov 20, 2024)10 benchmarksCoding 82Agents –Math –Science 82Reasoning 9010–
157Mistral Small 3.1 24B Instruct 2503open18.1 GB · 5 benchmarksCoding –Agents –Math –Science 81Reasoning –518.1 GB
158Llama 3.3 70B Instructopen51.3 GB · 12 benchmarksCoding 76Agents –Math –Science 81Reasoning 831251.3 GB
159Gemma 3 27B ITopen21.7 GB · 9 benchmarksCoding 70Agents –Math –Science 81Reasoning 74921.7 GB
160Claude 3.5 Haiku (Oct 22, 2024)8 benchmarksCoding 88Agents –Math –Science 73Reasoning –8–
161Gemma 3 12B ITopen10.9 GB · 7 benchmarksCoding –Agents –Math –Science 72Reasoning 66710.9 GB
162Phi 4 Mini Instructopen4.8 GB · 3 benchmarks—Coding –Agents –Math –Science 62Reasoning –34.8 GB
163Granite 4.0 Microopen3.7 GB · 3 benchmarks—Coding –Agents –Math –Science 62Reasoning –33.7 GB
164Granite 4.0 350Mopen1.0 GB · 3 benchmarks—Coding –Agents –Math –Science 58Reasoning –31.0 GB
165Llama 3.1 8B Instructopen7.9 GB · 9 benchmarksCoding 56Agents –Math –Science 56Reasoning 7097.9 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.