Coding
SciCode Leaderboard
SciCode asks a model to write real scientific-computing code — implementing multi-step numerical methods drawn from physics, chemistry, biology, and math research — and checks the output against reference solutions. It was built by a team of scientists and ML researchers to test coding ability on genuine research problems, not typical software tasks.
Source: epoch66 open models ranked+141 proprietaryData through Oct 2026
All models ranked on SciCode
Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.
| # | Model | Score | VRAM @ Q4 |
|---|---|---|---|
| 1 | Claude Opus 5.5 Max · proprietary | 66.9% | API only |
| 2 | Claude Opus 5.5 (xhigh) · proprietary | 65.0% | API only |
| 3 | Claude Fable 5.1 Max · proprietary | 63.1% | API only |
| 4 | Gemini 4 Argon (high) · proprietary | 61.8% | API only |
| 5 | Claude Fable 5 Max · proprietary | 61.0% | API only |
| 6 | Claude Sonnet 5.5 Max · proprietary | 61.0% | API only |
| 7 | Claude Fable 5.1 (xhigh) · proprietary | 60.9% | API only |
| 8 | MiMo V2.6 Pro RL · 1020B | 60.9% | ~613 GB |
| 9 | Claude Opus 5.5 (high) · proprietary | 60.4% | API only |
| 10 | Muse Wonder 1.0 Max · proprietary | 60.3% | API only |
| 11 | Gemini 3.7 Flash (medium) · proprietary | 59.8% | API only |
| 12 | Muse Spark 1.3 (xhigh) · proprietary | 59.7% | API only |
| 13 | Kimi K3 · 2779.9B | 59.5% | ~1674 GB |
| 14 | Claude Opus 5.5 (medium) · proprietary | 59.3% | API only |
| 15 | GLM 5.3 · 753.3B | 59.0% | ~456 GB |
| 16 | Step 5 Preview (unspecified) · proprietary | 58.9% | API only |
| 17 | Muse Spark 1.1 · proprietary | 58.8% | API only |
| 18 | Muse Spark 1.1 (xhigh) · proprietary | 58.8% | API only |
| 19 | Muse Spark 1.3 Max · proprietary | 58.8% | API only |
| 20 | Gemini 3.1 Pro Preview · proprietary | 58.7% | API only |
| 21 | Claude Opus 5.5 (low) · proprietary | 58.6% | API only |
| 22 | GPT 5.6 Sol (high) · proprietary | 57.8% | API only |
| 23 | Grok 4.7 (high) · proprietary | 57.8% | API only |
| 24 | Claude Fable 5.1 (high) · proprietary | 57.6% | API only |
| 25 | GPT 6 Sol Max · proprietary | 57.6% | API only |
| 26 | GPT 5.6 Sol (medium) · proprietary | 57.4% | API only |
| 27 | Grok 4.7 (xhigh) · proprietary | 57.4% | API only |
| 28 | Muse Spark 1.2 (xhigh) · proprietary | 57.4% | API only |
| 29 | Claude Sonnet 5.5 (xhigh) · proprietary | 57.3% | API only |
| 30 | Gemini 3.7 Flash (high) · proprietary | 57.2% | API only |
| 31 | GPT 5.6 Sol (xhigh) · proprietary | 57.1% | API only |
| 32 | GPT 5.6 Sol Max · proprietary | 57.1% | API only |
| 33 | Gemini 3.8 Flash (high) · proprietary | 56.6% | API only |
| 34 | GPT 5.4 (Mar 05, 2026, xhigh) · proprietary | 56.6% | API only |
| 35 | GPT 6 Astra Max · proprietary | 56.5% | API only |
| 36 | Grok 4.6 (high) · proprietary | 56.5% | API only |
| 37 | Claude Opus 5 Max · proprietary | 56.4% | API only |
| 38 | GPT 5.6 Sol (low) · proprietary | 56.4% | API only |
| 39 | GPT 5.5 (high) · proprietary | 56.1% | API only |
| 40 | Grok 4.6 (medium) · proprietary | 55.9% | API only |
| 41 | GPT 5.5 (xhigh) · proprietary | 55.8% | API only |
| 42 | GPT 6.1 Sol (high) · proprietary | 55.8% | API only |
| 43 | Claude Fable 5.1 (low) · proprietary | 55.7% | API only |
| 44 | GPT 6 Astra (xhigh) · proprietary | 55.7% | API only |
| 45 | GPT 6.1 Sol (xhigh) · proprietary | 55.7% | API only |
| 46 | GPT 6 Astra (high) · proprietary | 55.4% | API only |
| 47 | Claude Fable 5.1 (medium) · proprietary | 55.3% | API only |
| 48 | Gemini 3.8 Flash (medium) · proprietary | 55.1% | API only |
| 49 | GPT 6 Sol (xhigh) · proprietary | 55.1% | API only |
| 50 | Claude Haiku 5.5 Max · proprietary | 55.0% | API only |
| 51 | Claude Opus 5 (xhigh) · proprietary | 55.0% | API only |
| 52 | Gemini 3.8 Flash (low) · proprietary | 55.0% | API only |
| 53 | GPT 5.6 Terra Max · proprietary | 55.0% | API only |
| 54 | Grok 4.5 (high) · proprietary | 55.0% | API only |
| 55 | GPT 6 Sol (high) · proprietary | 54.9% | API only |
| 56 | Grok 4.7 (low) · proprietary | 54.9% | API only |
| 57 | GPT 6 Luna Max · proprietary | 54.6% | API only |
| 58 | Claude Opus 4.7 Max · proprietary | 54.5% | API only |
| 59 | GPT 5.5 (medium) · proprietary | 54.5% | API only |
| 60 | Claude Opus 5 (high) · proprietary | 54.3% | API only |
| 61 | Claude Sonnet 5 (high) · proprietary | 54.3% | API only |
| 62 | GPT 6 Astra (medium) · proprietary | 54.2% | API only |
| 63 | GPT 6.1 Sol Max · proprietary | 54.2% | API only |
| 64 | Mistral Large 4.0 (unspecified) · proprietary | 54.2% | API only |
| 65 | Claude Sonnet 5 (xhigh) · proprietary | 54.0% | API only |
| 66 | GPT 6 Astra (low) · proprietary | 54.0% | API only |
| 67 | Gemini 3.5 Flash (high) · proprietary | 53.9% | API only |
| 68 | GPT 6 Sol (medium) · proprietary | 53.8% | API only |
| 69 | Claude Sonnet 5.5 (high) · proprietary | 53.7% | API only |
| 70 | Claude Sonnet 5 Max · proprietary | 53.6% | API only |
| 71 | Gemini 3.7 Flash (low) · proprietary | 53.6% | API only |
| 72 | GPT 5.6 Luna Max · proprietary | 53.6% | API only |
| 73 | Claude Opus 4.8 Max · proprietary | 53.5% | API only |
| 74 | GPT 6 Astra None · proprietary | 53.5% | API only |
| 75 | Kimi K2.6 · 1026.9B | 53.5% | ~620 GB |
| 76 | Gemini 3.6 Flash (high) · proprietary | 53.4% | API only |
| 77 | GPT 6.1 Sol (low) · proprietary | 53.2% | API only |
| 78 | GPT 6.1 Sol (medium) · proprietary | 53.2% | API only |
| 79 | Qwen3.8 Max (unspecified) · proprietary | 53.2% | API only |
| 80 | Grok 4.6 (xhigh) · proprietary | 53.0% | API only |
| 81 | Claude Sonnet 5.5 (medium) · proprietary | 52.9% | API only |
| 82 | GPT 5.5 Instant · proprietary | 52.5% | API only |
| 83 | GPT 5.6 Terra (high) · proprietary | 52.4% | API only |
| 84 | GPT 5.6 Terra (xhigh) · proprietary | 52.3% | API only |
| 85 | GPT 5.4 Mini (Mar 17, 2026, xhigh) · proprietary | 52.1% | API only |
| 86 | Qwen3.8 Max (Sep 02, unspecified) · proprietary | 52.1% | API only |
| 87 | DeepSeek V4.1 Flash · 763.2B | 51.8% | ~458 GB |
| 88 | Claude Haiku 5.5 (xhigh) · proprietary | 51.7% | API only |
| 89 | GPT 6 Luna (xhigh) · proprietary | 51.7% | API only |
| 90 | Claude Sonnet 5 (medium) · proprietary | 51.6% | API only |
| 91 | GLM 5.3 Flash · 321.3B | 51.6% | ~195 GB |
| 92 | GPT 5.5 (low) · proprietary | 51.6% | API only |
| 93 | GPT 5.6 Luna (high) · proprietary | 51.6% | API only |
| 94 | Muse Spark · proprietary | 51.5% | API only |
| 95 | MiMo V2.6 Flash RL · 309B | 51.3% | ~186 GB |
| 96 | GLM 5.2 · 753.3B | 51.2% | ~456 GB |
| 97 | DeepSeek V4 Pro 0813 · 1650.5B | 51.0% | ~991 GB |
| 98 | GPT 6 Luna (medium) · proprietary | 50.9% | API only |
| 99 | Claude Opus 5 (medium) · proprietary | 50.7% | API only |
| 100 | GPT 5.6 Luna (xhigh) · proprietary | 50.5% | API only |
| 101 | GPT 5.6 Terra (medium) · proprietary | 50.5% | API only |
| 102 | GPT 6 Luna (high) · proprietary | 50.3% | API only |
| 103 | GPT 6 Sol (low) · proprietary | 50.2% | API only |
| 104 | Grok Build 0.1 · proprietary | 50.2% | API only |
| 105 | MiMo V2.5 Pro · 1023.2B | 50.2% | ~614 GB |
| 106 | Claude Sonnet 5 (low) · proprietary | 50.1% | API only |
| 107 | MiniMax M2.7 · 228.7B | 50.1% | ~138 GB |
| 108 | DeepSeek V4 Pro · 1598.8B | 50.0% | ~960 GB |
| 109 | DeepSeek V4 Flash 0731 · 304.2B | 49.9% | ~183 GB |
| 110 | GPT 5.6 Terra (low) · proprietary | 49.9% | API only |
| 111 | Inkling Small · 266.0B | 49.6% | ~160 GB |
| 112 | Qwen3.7 Max · proprietary | 49.5% | API only |
| 113 | Grok 4.6 (low) · proprietary | 49.4% | API only |
| 114 | Claude Haiku 5.5 (low) · proprietary | 49.2% | API only |
| 115 | Claude Sonnet 5.5 (low) · proprietary | 49.1% | API only |
| 116 | Claude Haiku 5.5 (medium) · proprietary | 49.0% | API only |
| 117 | Kimi K2.5 · 1026.9B | 49.0% | ~620 GB |
| 118 | Claude Haiku 5.5 (high) · proprietary | 48.7% | API only |
| 119 | Grok 4.3 (high) · proprietary | 48.3% | API only |
| 120 | Claude Opus 5 (low) · proprietary | 48.0% | API only |
| 121 | Kimi K2.7 Code · 1026.9B | 47.8% | ~620 GB |
| 122 | GPT 5.5 None · proprietary | 47.3% | API only |
| 123 | GPT 6 Sol None · proprietary | 47.3% | API only |
| 124 | GPT 5.4 Nano (Mar 17, 2026, xhigh) · proprietary | 47.2% | API only |
| 125 | GPT 5.6 Sol None · proprietary | 47.1% | API only |
| 126 | MiniMax M3 · 427.0B | 47.1% | ~257 GB |
| 127 | Inkling · 952.4B | 47.0% | ~572 GB |
| 128 | GPT 6 Luna (low) · proprietary | 46.9% | API only |
| 129 | Claude Sonnet 4.6 Max · proprietary | 46.8% | API only |
| 130 | GPT 5.6 Luna (medium) · proprietary | 46.8% | API only |
| 131 | Qwen3.8 27B · 27.8B | 46.6% | ~17 GB |
| 132 | Gemini 2.5 Pro · proprietary | 46.3% | API only |
| 133 | GPT 5.6 Luna (low) · proprietary | 46.1% | API only |
| 134 | Qwen3.7 Plus · proprietary | 46.1% | API only |
| 135 | Claude Sonnet 4.5 (Sep 29, 2025, unspecified) · proprietary | 45.7% | API only |
| 136 | Gemma 4 31B IT · 31.3B | 45.5% | ~20 GB |
| 137 | GLM 4.7 · 358.3B | 45.1% | ~216 GB |
| 138 | GPT 5.6 Terra None · proprietary | 45.1% | API only |
| 139 | Ring 2.6 1T · 1025.4B | 45.0% | ~2057 GB |
| 140 | DeepSeek V4 Flash · 290.9B | 44.9% | ~175 GB |
| 141 | Muse Glimmer 30B · 29.8B | 44.9% | ~18 GB |
| 142 | GLM 5.1 · 753.9B | 44.8% | ~457 GB |
| 143 | Gemini 3.1 Flash Lite · proprietary | 43.4% | API only |
| 144 | GPT 5.1 (Nov 13, 2025, unspecified) · proprietary | 43.3% | API only |
| 145 | GPT 6 Luna None · proprietary | 43.1% | API only |
| 146 | MiMo V2.5 · 310.8B | 43.1% | ~187 GB |
| 147 | GPT 5 (Aug 07, 2025, unspecified) · proprietary | 42.9% | API only |
| 148 | O3 Mini (Jan 31, 2025, high) · proprietary | 42.8% | API only |
| 149 | Nova 2.0 Pro Preview (medium) · proprietary | 42.7% | API only |
| 150 | Claude Haiku 4.5 (Oct 01, 2025, unspecified) · proprietary | 42.3% | API only |
| 151 | Qwen3 235B A22B Thinking 2507 · 235.1B | 41.4% | ~142 GB |
| 152 | Gemini 3.5 Flash Lite · proprietary | 41.3% | API only |
| 153 | Cogito 671B V2.1 · 671B | 41.0% | ~406 GB |
| 154 | Qwen3.6 Plus · proprietary | 40.7% | API only |
| 155 | GPT 4.1 Mini (Apr 14, 2025) · proprietary | 40.4% | API only |
| 156 | GPT 5.6 Luna None · proprietary | 40.4% | API only |
| 157 | NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 560.5B | 40.3% | ~370 GB |
| 158 | Mistral Medium 2604 · proprietary | 40.2% | API only |
| 159 | Claude Sonnet 4 (May 14, 2025, unspecified) · proprietary | 40.1% | API only |
| 160 | Gemma 4 26B A4B IT · 25.8B | 40.1% | ~16 GB |
| 161 | Step 3.7 Flash · 201.4B | 40.1% | ~133 GB |
| 162 | GPT 5 Mini (Aug 07, 2025, unspecified) · proprietary | 39.2% | API only |
| 163 | Magistral Medium 2509 · proprietary | 39.2% | API only |
| 164 | DeepSeek v3 0324 · 684.5B | 39.0% | ~415 GB |
| 165 | GPT 5 Mini (Aug 07, 2025, high) · proprietary | 39.0% | API only |
| 166 | DeepSeek V3.2 Exp · 685.4B | 38.9% | ~415 GB |
| 167 | Mercury 2 · proprietary | 38.7% | API only |
| 168 | Nova 2.0 Pro Preview (low) · proprietary | 38.7% | API only |
| 169 | Command A Plus 05 2026 BF16 · 218.8B | 38.5% | ~132 GB |
| 170 | Mercury 2.5 (unspecified) · proprietary | 38.5% | API only |
| 171 | GLM 4.6 · 356.8B | 38.4% | ~215 GB |
| 172 | DeepSeek R1 · 684.5B | 38.3% | ~415 GB |
| 173 | DeepSeek V3.1 Terminus · 684.5B | 38.0% | ~415 GB |
| 174 | Qwen3.6 27B · 27.8B | 37.3% | ~17 GB |
| 175 | Mistral Large 3 675B Instruct 2512 · 675B | 36.6% | ~446 GB |
| 176 | Qwen3.6 35B A3B · 36.0B | 36.6% | ~22 GB |
| 177 | Trinity Large Thinking · 398.6B | 36.1% | ~798 GB |
| 178 | GPT OSS 120B · 116.8B | 36.0% | ~71 GB |
| 179 | NVIDIA Nemotron 3 Super 120B A12B BF16 · 123.6B | 36.0% | ~75 GB |
| 180 | DeepSeek v3 · 684.5B | 35.8% | ~415 GB |
| 181 | Qwen3.5 122B A10B · 125.1B | 35.6% | ~75 GB |
| 182 | Qwen3 32B · 32.8B | 35.4% | ~20 GB |
| 183 | Magistral Small 2509 · 24.0B | 35.2% | ~15 GB |
| 184 | GPT OSS 20B · 20.9B | 34.4% | ~13 GB |
| 185 | Llama 4 Maverick 17B 128E Instruct · 401.6B | 33.1% | ~265 GB |
| 186 | Qwen3 30B A3B Thinking 2507 · 30.5B | 33.0% | ~19 GB |
| 187 | Mistral Medium 2508 · proprietary | 32.4% | API only |
| 188 | Qwen3 Coder Next · 79.7B | 32.3% | ~48 GB |
| 189 | Qwen3 14B · 14.8B | 31.6% | ~9.5 GB |
| 190 | Qwen3.5 9B · 9.7B | 29.5% | ~6.4 GB |
| 191 | Qwen3.5 35B A3B · 36.0B | 29.3% | ~22 GB |
| 192 | Devstral Small 2 24B Instruct 2512 · 24.0B | 28.8% | ~15 GB |
| 193 | Nova 2.0 Pro Preview None · proprietary | 28.1% | API only |
| 194 | Mistral Small 3.1 24B Instruct 2503 · 24.0B | 27.8% | ~15 GB |
| 195 | Claude 3.5 Haiku (Oct 22, 2024) · proprietary | 27.4% | API only |
| 196 | Mistral Small 3.2 24B Instruct 2506 · 24.0B | 26.4% | ~15 GB |
| 197 | Llama 3.3 70B Instruct · 70.6B | 26.0% | ~47 GB |
| 198 | GPT 4.1 Nano (Apr 14, 2025) · proprietary | 25.9% | API only |
| 199 | MiMo v2 Flash · 309.8B | 25.9% | ~186 GB |
| 200 | Granite 4.1 30B · 28.9B | 25.8% | ~18 GB |
| 201 | Solar Pro 3 · proprietary | 25.5% | API only |
| 202 | Gemma 3 27B IT · 27.4B | 23.3% | ~18 GB |
| 203 | Qwen3 8B · 8.2B | 22.6% | ~5.5 GB |
| 204 | Llama 4 Scout 17B 16E Instruct · 108.6B | 17.0% | ~72 GB |
| 205 | Gemma 3 12B IT · 12.2B | 16.4% | ~8.0 GB |
| 206 | Llama 3.1 8B Instruct · 8.0B | 13.2% | ~5.3 GB |
| 207 | Phi 4 Mini Instruct · 3.8B | 10.8% | ~2.9 GB |
Score vs model size
Which models give the most quality for their size — the ones worth running locally.
- Phi 4 Mini Instruct, 4B, score 10.8% — on the efficiency frontier (best score at its size or smaller).
- Llama 3.1 8B Instruct, 8B, score 13.2% — on the efficiency frontier (best score at its size or smaller).
- Qwen3 8B, 8B, score 22.6% — on the efficiency frontier (best score at its size or smaller).
- Qwen3.5 9B, 10B, score 29.5% — on the efficiency frontier (best score at its size or smaller).
- Qwen3 14B, 15B, score 31.6% — on the efficiency frontier (best score at its size or smaller).
- GPT OSS 20B, 21B, score 34.4% — on the efficiency frontier (best score at its size or smaller).
- Magistral Small 2509, 24B, score 35.2% — on the efficiency frontier (best score at its size or smaller).
- Gemma 4 26B A4B IT, 26B, score 40.1% — on the efficiency frontier (best score at its size or smaller).
- Qwen3.8 27B, 28B, score 46.6% — on the efficiency frontier (best score at its size or smaller).
- MiniMax M2.7, 229B, score 50.1% — on the efficiency frontier (best score at its size or smaller).
- MiMo V2.6 Flash RL, 309B, score 51.3% — on the efficiency frontier (best score at its size or smaller).
- GLM 5.3 Flash, 321B, score 51.6% — on the efficiency frontier (best score at its size or smaller).
- GLM 5.3, 753B, score 59.0% — on the efficiency frontier (best score at its size or smaller).
- MiMo V2.6 Pro RL, 1T, score 60.9% — on the efficiency frontier (best score at its size or smaller).
SciCode: frequently asked questions
- What is the best open LLM on SciCode?
- MiMo V2.6 Pro RL is the top open model on SciCode, scoring 60.9%. Among all models tested — including proprietary ones — it ranks #7. The top model overall is Claude Opus 5.5 Max (Anthropic) at 66.9%.
- What's the best SciCode model you can run on a 24 GB GPU?
- Qwen3.8 27B is the highest-scoring open model that fits in 24 GB at 4-bit quantization (~17 GB), scoring 46.6% on SciCode.
- What's the best SciCode model you can run on a 12 GB GPU?
- Qwen3 14B is the highest-scoring open model that fits in 12 GB at 4-bit quantization (~9.5 GB), scoring 31.6% on SciCode.
- Can open models match proprietary models on SciCode?
- Not quite on SciCode: the strongest proprietary model (Claude Opus 5.5 Max) scores 66.9%, ahead of the best open model (MiMo V2.6 Pro RL) at 60.9% — but you can run the open one yourself.
Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.