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.
40 models · 40 results · data as of 2026-10-29 · hub.harborframework.com ↗
Leaderboard
GPT-5.5 xhigh
All results
40 rows| # | Model | Effort | Terminal-Bench 2.0 | Harness |
|---|---|---|---|---|
| 1 | GPT-5.5 | xhigh | 84.7% ±2.5 | clnkr |
| 2 | Opus 4.7 | max | 80.2% ±2.1 | WOZCODE |
| 3 | Gemini 3.1 Pro | high | 80.2% ±4.1 | Gemini CLI |
| 4 | GPT-5.3 Codex | xhigh | 78.4% ±2.7 | Terminus 2 |
| 5 | Opus 4.6 | max | 76.4% ±2.9 | Claude Code |
| 6 | Gemini 3 Pro | high | 69.4% ±3.0 | Letta Code |
| 7 | GPT-5.2 Codex | xhigh | 66.5% ±3.1 | Deep Agents |
| 8 | GPT-5.2 | xhigh | 64.9% ±2.9 | Terminus 2 |
| 9 | Gemini 3 Flash | high | 64.3% ±3.0 | Gemini CLI |
| 10 | Opus 4.5 | high | 63.1% | OpenCode |
| 11 | GPT-5.1 Codex Mini | high | 61.6% ±3.0 | Crux |
| 12 | GPT-5.1 Codex | max | 60.4% ±3.2 | Terminus 2 |
| 13 | Grok 4.20 | xhigh | 57.3% | Grok CLI |
| 14 | Sonnet 4.6 | max | 53.4% ±2.8 | Simplai Agent |
| 15 | GLM 5 | high | 52.4% ±2.6 | Terminus 2 |
| 16 | GPT-5 | high | 49.6% ±2.9 | Mini-SWE-Agent |
| 17 | GPT-5.1 | high | 47.6% ±2.8 | Terminus 2 |
| 18 | Sonnet 4.5 | high | 46.5% ±2.9 | Claude Code |
| 19 | MiniMax M2.7 | high | 45.1% ±2.9 | Harness Agent |
| 20 | GPT-5 Codex | high | 44.3% ±2.8 | Mini-SWE-Agent |
| 21 | Kimi K2.5 | max | 43.2% ±2.9 | Terminus 2 |
| 22 | MiniMax M2.5 | high | 42.7% ±2.6 | Terminus 2 |
| 23 | DeepSeek V3.2 | high | 39.6% ±2.8 | Terminus 2 |
| 24 | Opus 4.1 | high | 38.0% ±2.9 | Claude Code |
| 25 | MiniMax M2.1 | 36.6% ±2.9 | Terminus 2 | |
| 26 | Kimi K2 | medium | 35.7% ±2.7 | OpenHands |
| 27 | Haiku 4.5 | xhigh | 35.5% ±2.7 | OpenHands |
| 28 | GPT-5 Mini | high | 34.8% ±2.6 | Mini-SWE-Agent |
| 29 | GLM 4.7 | none | 33.4% ±2.5 | Crux |
| 30 | Gemini 2.5 Pro | high | 32.6% ±2.8 | OpenHands |