Benchmarks / Agentic
Terminal-Bench 2.0
Terminal-Bench 2.0: task resolution rate for agent+model pairs on 89 terminal tasks. Kept for time travel.
Terminal-Bench 2.0
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-5.5 | xhigh | 84.7% |
| 2 | Gemini 3.1 Pro | high | 80.2% |
| 3 | Opus 4.7 | max | 80.2% |
| 4 | GPT-5.3 Codex | xhigh | 78.4% |
| 5 | Opus 4.6 | max | 76.4% |
| 6 | Gemini 3 Pro | high | 69.4% |
| 7 | GPT-5.2 Codex | xhigh | 66.5% |
| 8 | GPT-5.2 | xhigh | 64.9% |
| 9 | Gemini 3 Flash | high | 64.3% |
| 10 | Opus 4.5 | high | 63.1% |
| 11 | GPT-5.1 Codex Mini | high | 61.6% |
| 12 | GPT-5.1 Codex | max | 60.4% |
| 13 | Grok 4.20 | xhigh | 57.3% |
| 14 | Sonnet 4.6 | max | 53.4% |
| 15 | GLM 5 | high | 52.4% |
| 16 | GPT-5 | high | 49.6% |
| 17 | GPT-5.1 | high | 47.6% |
| 18 | Sonnet 4.5 | high | 46.5% |
| 19 | MiniMax M2.7 | high | 45.1% |
| 20 | GPT-5 Codex | high | 44.3% |
| 21 | Kimi K2.5 | max | 43.2% |
| 22 | MiniMax M2.5 | high | 42.7% |
| 23 | DeepSeek V3.2 | high | 39.6% |
| 24 | Opus 4.1 | high | 38.0% |
| 25 | MiniMax M2.1 | – | 36.6% |
| 26 | Kimi K2 | medium | 35.7% |
| 27 | Haiku 4.5 | xhigh | 35.5% |
| 28 | GPT-5 Mini | high | 34.8% |
| 29 | GLM 4.7 | none | 33.4% |
| 30 | Gemini 2.5 Pro | high | 32.6% |
| 31 | MiniMax M2 | – | 30.0% |
| 32 | Grok 4 | high | 27.2% |
| 33 | Qwen 3 Coder | medium | 27.2% |
| 34 | Qwen 3.6 35B-A3B | max | 24.6% |
| 35 | GLM 4.6 | none | 24.5% |
| 36 | GPT-5 Nano | high | 21.8% |
| 37 | gpt-oss-120b | high | 18.7% |
| 38 | Gemini 2.5 Flash | medium | 17.1% |
| 39 | gpt-oss-20b | high | 17.0% |
| 40 | Qwen 3.5 9B | low | 9.2% |