Terminal-Bench v4.0
Tests whether an agent can complete complex work through the terminal.
Leaderboard
| Rank | Model | Score |
|---|---|---|
| #1 | Muse Wonder (Full) | 61.4 |
| #2 | Claude Opus 5.5 (max) | 60 |
| #3 | Mistral Large 4 (Base (Preview)) | 59.8 |
| #4 | GPT-6 Astra (max) | 59.1 |
| #5 | Muse Wonder (Standard (API)) | 58.2 |
| #6 | Claude Fable 5.1 (max with fallback) | 52 |
| #7 | Claude Sonnet 5.5 (max) | 50.5 |
| #8 | GPT-5.6 Sol (max) | 39.9 |