# Terminal-Bench

Tests whether AI agents can complete real tasks by typing commands in a computer terminal, like a developer or system administrator would.

Category: coding. Type: leaderboard. Status: active.
Metric: Share of Terminal-Bench 4.1 tasks completed (%) with 95% CI (higher is better).
Source: Terminal-Bench (Stanford / Harbor / Laude Institute) — https://www.tbench.ai/leaderboard, retrieved 2026-11-17, license: not stated.

| Rank | Model | Agent | Resolution rate | Cost |
|---|---|---|---|---|
| 1 | Opus 5.5 ( max ) | Claude Code | 65.8% ± 3.1% | $4.7k |
| 1 | Fable 5.2 ( max ) | Claude Code | 65.8% ± 2.8% | $16.5k |
| 3 | Gemini 4 Carbon ( high ) | Antigravity CLI | 64.5% ± 3.2% | $1.4k |
| 4 | Sonnet 5.5 ( max ) | Claude Code | 63.9% ± 3.0% | $7.3k |
| 5 | GPT-6.1 Sol ( max ) | Codex | 60.9% ± 3.0% | $0.6k |
| 6 | GPT-6 Astra ( max ) | Codex | 60.3% ± 2.8% | $3.3k |
| 7 | Kimi K3.1 Pro ( max ) | Claude Code | 59.7% ± 4.2% | $1.1k |
| 8 | Fable 5.1 ( xhigh ) | Claude Code | 57.6% ± 3.3% | $4.9k |
| 9 | Gemini 4 Hydrogen ( high ) | Antigravity CLI | 56.4% ± 3.0% | $0.8k |
| 10 | Gemini 4 Argon ( high ) | Antigravity CLI | 55.5% ± 2.8% | $1.1k |
| 11 | Opus 5 ( xhigh ) | Claude Code | 53.3% ± 3.1% | $6.1k |
| 11 | GPT-6.1 Terra ( max ) | Codex | 53.3% ± 3.1% | $0.9k |
| 13 | GPT-6 Sol ( max ) | Codex | 50.3% ± 3.2% | $1.5k |
| 14 | Muse Wonder ( xhigh ) | Muse Code | 47.0% ± 3.8% | $2.5k |
| 15 | Fable 5 ( max ) | Claude Code | 44.8% ± 3.8% | $7.3k |
| 16 | GLM-5.3 ( max ) | Claude Code | 41.2% ± 3.3% | $2.7k |
| 17 | Haiku 5.5 ( max ) | Claude Code | 40.0% ± 3.6% | $1.0k |
| 18 | Grok 4.7 ( xhigh ) | Grok Build | 38.8% ± 3.6% | $3.7k |
| 19 | GPT-5.6 Sol ( max ) | Codex | 38.5% ± 3.8% | $2.5k |
| 19 | DeepSeek V4.1 Flash 1017 ( max ) | Claude Code | 38.5% ± 3.8% | $0.2k |
| 21 | Hy4 ( max ) | Claude Code | 37.9% ± 3.5% | $0.3k |
| 22 | GLM-5.3-Flash ( max ) | Claude Code | 37.0% ± 3.6% | $0.4k |
| 23 | Muse Ember ( xhigh ) | Muse Code | 30.6% ± 3.8% | $0.2k |
| 24 | Mistral Large 4 ( max ) | Claude Code | 27.6% ± 3.9% | $0.7k |
| 25 | Qwen3.8-Max-0902 ( max ) | Claude Code | 27.0% ± 3.9% | $4.6k |

Notes: Terminal-Bench 4.1 leaderboard; as-of date not shown on page (latest listed release Nov 12, 2026).
