Terminal-Bench 4
Hard tasks done entirely in a Linux terminal: building, debugging, data work, system administration. Official Terminal-Bench 4.x leaderboard; each entry is a model in a named agent harness.
| Category | Agents |
|---|---|
| Metric | Accuracy (%), 5 trials |
| Source | Terminal-Bench (original) |
| License | Apache 2.0 |
| Benchmark page | www.tbench.ai |
| Status | Scored |
| Weight | 1.2 within Agents |
| Fit | Discrimination 2.51, noise 0.27 |
Results
| # | Model | Provider | Result | Date |
|---|---|---|---|---|
| 1 | Claude Opus 5.5 Max | Anthropic | 65.8% | 2026-09-22 |
| 2 | Claude Sonnet 5.5 Max | Anthropic | 63.9% | 2026-09-28 |
| 3 | GPT-6.1 Sol Max | OpenAI | 60.9% | 2026-09-29 |
| 4 | GPT-6 Astra Max | OpenAI | 60.3% | 2026-09-03 |
| 5 | GPT-6 Astra High | OpenAI | 60.0% | 2026-09-03 |
| 6 | GPT-6 Astra Extra High | OpenAI | 59.7% | 2026-09-03 |
| 7 | Claude Fable 5.1 Extra High | Anthropic | 57.6% | 2026-09-01 |
| 8 | Claude Fable 5.1 Max | Anthropic | 57.3% | 2026-09-01 |
| 9 | GPT-6 Astra Medium | OpenAI | 55.2% | 2026-09-03 |
| 10 | Claude Fable 5.1 High | Anthropic | 54.9% | 2026-09-01 |
| 11 | Claude Fable 5.1 Medium | Anthropic | 53.9% | 2026-09-01 |
| 12 | Claude Opus 5 Extra High | Anthropic | 53.3% | 2026-07-24 |
| 13 | Claude Opus 5 Max | Anthropic | 52.1% | 2026-07-24 |
| 14 | GPT-6 Astra Low | OpenAI | 50.9% | 2026-09-03 |
| 15 | GPT-6 Sol Max | OpenAI | 50.3% | 2026-09-22 |
| 16 | Claude Opus 5 High | Anthropic | 49.7% | 2026-07-24 |
| 17 | Claude Opus 5 Medium | Anthropic | 45.2% | 2026-07-24 |
| 18 | Claude Fable 5 Max | Anthropic | 44.9% | 2026-06-09 |
| 19 | Claude Fable 5.1 Low | Anthropic | 43.3% | 2026-09-01 |
| 20 | GLM-5.3 Max | Zhipu AI | 41.2% | 2026-08-14 |
| 21 | Grok 4.7 Extra High | xAI | 38.8% | 2026-09-21 |
| 22 | GPT-5.6 Sol Max | OpenAI | 38.5% | 2026-06-26 |
| 23 | GLM-5.3-Flash Non-reasoning | Zhipu AI | 35.5% | 2026-08-26 |
| 24 | Claude Opus 5 Low | Anthropic | 34.9% | 2026-07-24 |
| 25 | Qwen3.8 Max 0902 Max | Alibaba | 27.0% | 2026-09-02 |
| 26 | Claude Opus 4.8 Max | Anthropic | 23.6% | 2026-05-28 |
| 27 | GPT-5.6 Terra Max | OpenAI | 21.5% | 2026-06-26 |
| 28 | Grok 4.6 High | xAI | 19.7% | 2026-08-12 |
| 29 | Gemini 3.8 Flash High | 18.2% | 2026-09-02 | |
| 30 | GPT-5.6 Luna Max | OpenAI | 17.6% | 2026-06-26 |
| 31 | GPT-6 Luna Max | OpenAI | 15.2% | 2026-09-22 |
| 32 | Muse Spark 1.3 Extra High | Meta | 14.9% | 2026-09-02 |
| 33 | Gemini 3.7 Flash High | 12.4% | 2026-08-13 | |
| 34 | Grok 4.5 High | xAI | 12.4% | 2026-07-16 |
| 35 | Claude Sonnet 5 Max | Anthropic | 12.1% | 2026-06-30 |