Benchmarks · agentic
Terminal-Bench 2.1
Terminal-Bench 2.1 (verified revision of 2.0): task resolution rate for agent+model pairs on terminal tasks. Kept for time travel.
13 models · 18 results · data as of 2026-10-29 · hub.harborframework.com ↗
Leaderboard
GPT-6 Astra high
All results
18 rows| # | Model | Effort | Terminal-Bench 2.1 | Harness |
|---|---|---|---|---|
| 1 | GPT-6 Astra | high | 87.4% ±1.8 | Codex |
| 2 | GPT-6 Astra | medium | 87.0% ±1.9 | Codex |
| 3 | GPT-6 Astra | low | 86.7% ±1.8 | Codex |
| 4 | GPT-6 Astra | max | 86.7% ±1.4 | Codex |
| 5 | GPT-6 Astra | xhigh | 85.8% ±1.4 | Codex |
| 6 | Fable 5 | xhigh | 83.8% ±2.3 | Claude Code |
| 7 | GPT-5.5 | xhigh | 83.2% ±2.4 | Terminus 2 |
| 8 | Fable 5 | high | 80.5% ±2.3 | Terminus 2 |
| 9 | Grok 4.5 | high | 79.3% ±2.9 | Cursor CLI |
| 10 | Opus 4.8 | high | 78.9% ±2.6 | Claude Code |
| 11 | GPT-5.6 Terra | max | 78.4% ±2.5 | Codex |
| 12 | Muse Spark 1.1 | xhigh | 76.2% ±2.4 | mini-SWE-agent |
| 13 | GPT-5.6 Luna | max | 75.7% ±2.6 | Codex |
| 14 | Sonnet 5 | high | 74.6% ±3.2 | Claude Code |
| 15 | Gemini 3 Pro | high | 73.9% ±2.7 | Gemini CLI |
| 16 | Opus 4.7 | max | 68.9% ±2.7 | Terminus 2 |
| 17 | Gemini 3.1 Pro | high | 65.8% ±3.2 | Terminus 2 |
| 18 | GLM 5.1 | max | 58.6% ±2.4 | Claude Code |