Terminal-Bench
Command-line agent tasks; best result per model across harnesses.
- Published by
- Terminal-Benchdata via Epoch AI Benchmarking Hub
- Category
- Agents & tools
- Index weight
- 1.0
- Models
- 66
- Data as of
- 9 Sept 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
- 1GPT-5.584.7%
- 2GPT-5.481.8%
- 3Claude Opus 4.780.2%
- 4Gemini 3.1 Pro80.2%
- 5Claude Opus 4.679.8%
- 6GPT-5.3 Codex78.4%
- 7Gemini 3 Pro69.4%
- 8GPT-5.2 Codex66.5%
- 9GPT-5.264.9%
- 10GPT-5.264.9%
- 11Gemini 3 Flash64.3%
- 12Claude Opus 4.563.1%
- 13GPT-5.1-Codex-mini61.6%
- 14GPT-5.1-Codex-Max60.4%
- 15GPT-6 Astra58.2%
66 of 66
| # | |||||
|---|---|---|---|---|---|
| 1 |  GPT-5.5OpenAI | 84.7% | – | 66.6 | – |
| 2 | GPT-5.4OpenAI | 81.8% | – | 63.4 | – |
| 3 |  Claude Opus 4.7Anthropic | 80.2% | – | 65.5 | – |
| 4 |  Gemini 3.1 ProGoogle | 80.2% | – | 63.9 | – |
| 5 | Claude Opus 4.6Anthropic | 79.8% | – | 63.1 | – |
| 6 | GPT-5.3 CodexOpenAI | 78.4% | – | 62.4 | – |
| 7 | Gemini 3 ProGoogle | 69.4% | – | 61.6 | – |
| 8 | GPT-5.2 CodexOpenAI | 66.5% | – | 60.7 | – |
| 9 | GPT-5.2OpenAI | 64.9% | – | 61.5 | – |
| 10 | GPT-5.2OpenAI | 64.9% | – | 58.5 | – |
| 11 | Gemini 3 FlashGoogle | 64.3% | – | 56.6 | – |
| 12 | Claude Opus 4.5Anthropic | 63.1% | – | 58.0 | – |
| 13 | GPT-5.1-Codex-miniOpenAI | 61.6% | – | 53.4 | – |
| 14 | GPT-5.1-Codex-MaxOpenAI | 60.4% | – | – | – |
| 15 | GPT-6 AstraOpenAI | 58.2% | – | 70.6 | – |
| 16 | Claude Fable 5.1Anthropic | 57.9% | – | 70.2 | – |
| 17 | GPT-6 AstraOpenAI | 57.9% | – | 69.5 | – |
| 18 | GPT-6 AstraOpenAI | 57.9% | – | 68.6 | – |
| 19 | GPT-5.1-CodexOpenAI | 57.8% | – | 57.1 | – |
| 20 |  Grok 4.20xAI | 57.3% | – | 57.9 | – |
| 21 | GPT-6 AstraOpenAI | 54.2% | – | 67.4 | – |
| 22 | Claude Sonnet 4.6Anthropic | 53.4% | – | 60.3 | – |
| 23 |  GLM-5Zhipu AIopen | 52.4% | via Terminus 2 | 59.0 | – |
| 24 | Claude Opus 5Anthropic | 51.8% | – | 66.8 | – |
| 25 | GPT-6 AstraOpenAI | 50.6% | – | 65.8 | – |
| 26 | GPT-5OpenAI | 49.6% | – | 59.5 | – |
| 27 | GPT-5OpenAI | 49.6% | – | 58.5 | – |
| 28 | GPT-5.1OpenAI | 47.6% | dated, via Terminus 2 | 59.4 | – |
| 29 | GPT-5.1OpenAI | 47.6% | dated, medium, via Terminus 2 | 52.8 | – |
| 30 | Claude Sonnet 4.5Anthropic | 46.5% | – | 52.3 | – |
| 31 |  MiniMax-M2.7MiniMaxopen | 45.1% | – | 56.5 | – |
| 32 | Claude Fable 5Anthropic | 44.5% | – | 65.8 | – |
| 33 | GPT-5-CodexOpenAI | 44.3% | – | 58.1 | – |
| 34 |  Kimi K2.5Moonshot AIopen | 43.2% | via Terminus 2 | 56.1 | – |
| 35 | MiniMax-M2.5MiniMaxopen | 42.7% | – | 53.5 | – |
| 36 | GLM-5.3Zhipu AIopen | 41.8% | – | 59.9 | – |
| 37 | Claude Opus 4.1Anthropic | 38.0% | via Terminus 2 | 52.6 | – |
| 38 | GPT-5.6 SolOpenAI | 37.3% | – | 65.5 | – |
| 39 | MiniMax-M2MiniMaxopen | 36.6% | – | 55.1 | – |
| 40 | Kimi K2 ThinkingMoonshot AIopen | 35.7% | thinking, via Terminus 2 | 52.8 | – |
| 41 | Claude Haiku 4.5Anthropic | 35.5% | – | 47.9 | – |
| 42 | GPT-5 miniOpenAI | 34.8% | – | 55.8 | – |
| 43 | GLM-4.7Zhipu AIopen | 33.4% | via Terminus 2 | 54.9 | – |
| 44 | Gemini 2.5 ProGoogle | 32.6% | via Terminus 2 | 54.1 | – |
| 45 | GPT-5 miniOpenAI | 31.9% | – | 54.5 | – |
| 46 | Kimi K2Moonshot AIopen | 27.8% | via Terminus 2 | 48.8 | – |
| 47 | Grok 4xAI | 27.2% | 0709, via OpenHands | 57.2 | – |
| 48 |  Qwen3 Coder 480BAlibabaopen | 27.2% | – | 48.5 | – |
| 49 | Grok Code Fast 1xAI | 25.8% | – | 45.9 | – |
| 50 | GLM-4.6Zhipu AIopen | 24.5% | via Terminus 2 | 47.6 | – |
| 51 | Claude Opus 4.8Anthropic | 23.6% | – | 59.4 | – |
| 52 | Qwen3.6 35B-A3BAlibabaopen | 23.0% | – | 53.2 | – |
| 53 | GPT-5 nanoOpenAI | 21.8% | – | 51.1 | – |
| 54 | GPT-5.6 TerraOpenAI | 21.5% | – | 58.0 | – |
| 55 | Grok 4.6xAI | 20.3% | – | 58.7 | – |
| 56 | Gemini 3.8 FlashGoogle | 19.1% | – | 60.1 | – |
| 57 | gpt-oss-120bOpenAIopen | 18.7% | via Terminus 2 | 47.6 | – |
| 58 | GPT-5.6 LunaOpenAI | 17.3% | – | 56.5 | – |
| 59 | Gemini 2.5 Flash 09Google | 17.1% | – | 50.7 | – |
| 60 | Gemini 2.5 FlashGoogle | 17.1% | – | 48.8 | – |