Terminal-Bench-Science
Research workflow tasks from practicing scientists, done in a terminal. Run by Vals AI.
| Category | Agents |
|---|---|
| Metric | Accuracy (%) |
| Source | Vals AI (evaluator) |
| License | None stated |
| Benchmark page | www.vals.ai |
| Status | Scored |
| Weight | 0.8 within Agents |
| Fit | Discrimination 3.12, noise 0.63 |
Results
| # | Model | Provider | Result | Date |
|---|---|---|---|---|
| 1 | GPT-6 Astra Max | OpenAI | 62.9% | 2026-11-09 |
| 2 | Claude Fable 5.2 High | Anthropic | 55.7% | 2026-11-09 |
| 3 | GPT-6.1 Sol Max | OpenAI | 52.9% | 2026-11-09 |
| 4 | Claude Sonnet 5.5 Max | Anthropic | 47.1% | 2026-11-09 |
| 5 | Claude Opus 5.5 Max | Anthropic | 47.1% | 2026-11-09 |
| 6 | Gemini 4 Argon High | 44.3% | 2026-11-09 | |
| 7 | Claude Fable 5.1 Max | Anthropic | 40.0% | 2026-11-09 |
| 8 | Muse Wonder Full Max | Meta | 37.1% | 2026-11-09 |
| 9 | Kimi K3.1 Pro Max | Moonshot AI | 31.4% | 2026-11-09 |
| 10 | GPT-6 Sol Max | OpenAI | 30.0% | 2026-11-09 |
| 11 | Claude Opus 5 Max | Anthropic | 27.1% | 2026-11-09 |
| 12 | Muse Wonder Max | Meta | 22.9% | 2026-11-09 |
| 13 | GPT-5.6 Sol Max | OpenAI | 20.0% | 2026-11-09 |
| 14 | Claude Fable 5 Max | Anthropic | 15.7% | 2026-11-09 |
| 15 | Qwen3.8 Max | Alibaba | 12.9% | 2026-11-09 |
| 16 | Grok 4.6 High | xAI | 11.4% | 2026-11-09 |
| 17 | Claude Haiku 5.5 Max | Anthropic | 10.0% | 2026-11-09 |
| 18 | Grok 4.7 Extra High | xAI | 10.0% | 2026-11-09 |
| 19 | GPT-5.6 Terra Max | OpenAI | 10.0% | 2026-11-09 |
| 20 | Muse Spark 1.3 Max | Meta | 10.0% | 2026-11-09 |
| 21 | Gemini 3.8 Flash High | 8.6% | 2026-11-09 | |
| 22 | Gemini 3.7 Flash High | 7.1% | 2026-11-09 | |
| 23 | Ember-1 | Fireworks | 5.7% | 2026-11-09 |
| 24 | GLM-5.3 Max | Zhipu AI | 5.7% | 2026-11-09 |
| 25 | Claude Sonnet 5 Max | Anthropic | 5.7% | 2026-11-09 |
| 26 | Gemini 3.5 Flash High | 5.7% | 2026-11-09 | |
| 27 | MiMo-V2.6-Flash | Xiaomi | 4.3% | 2026-11-09 |
| 28 | GPT-6 Luna Max | OpenAI | 4.3% | 2026-11-09 |
| 29 | DeepSeek V4 Pro 0813 Max | DeepSeek | 4.3% | 2026-11-09 |
| 30 | Gemini 3.6 Flash High | 4.3% | 2026-11-09 | |
| 31 | Claude Opus 4.8 Max | Anthropic | 4.3% | 2026-11-09 |
| 32 | Muse Spark 1.3 Extra High | Meta | 4.3% | 2026-11-09 |
| 33 | DeepSeek V4 Flash 0731 High | DeepSeek | 4.3% | 2026-11-09 |
| 34 | Step 5 Preview | StepFun | 2.9% | 2026-11-09 |
| 35 | Inkling 2 Small | Thinking Machines | 2.9% | 2026-11-09 |
| 36 | MiMo-V2.6-Pro | Xiaomi | 2.9% | 2026-11-09 |
| 37 | GLM-5.3-Flash Max | Zhipu AI | 2.9% | 2026-11-09 |
| 38 | Hy4 preview | Tencent | 1.4% | 2026-11-09 |
| 39 | Beam 1.0 | Reflection | 1.4% | 2026-11-09 |
| 40 | Gemini 3.1 Pro Preview High | 1.4% | 2026-11-09 | |
| 41 | Kimi K3 Max | Moonshot AI | 1.4% | 2026-11-09 |
| 42 | DeepSeek V4.1 Flash Max | DeepSeek | 0.0% | 2026-11-09 |
| 43 | GPT-5.6 Luna Max | OpenAI | 0.0% | 2026-11-09 |
| 44 | Qwen3.8 27B Extra High | Alibaba | 0.0% | 2026-11-09 |
| 45 | Mercury 2.5 High | Inception | 0.0% | 2026-11-09 |