Terminal-Bench 2.0 — leaderboard
89 terminal/CLI tasks testing model ability to use command-line tools, navigate filesystems, and automate workflows. Scored by accuracy across agent-model pairings.
Metric: Accuracy (%). Source: www.tbench.ai. Status: saturation imminent. 59 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 84.7 |
| 2 | GPT-5.4 | 81.8 |
| 3 | Gemini 3.1 Pro (Preview) | 80.2 |
| 4 | GPT-5.3 Codex | 78.4 |
| 5 | Claude Opus 4.6 | 69.9 |
| 6 | Gemini 3 Pro (Preview) | 69.4 |
| 7 | GPT-5.2 Codex | 66.5 |
| 8 | GPT-5.2 | 64.9 |
| 9 | GPT-5.2 (Medium) | 64.9 |
| 10 | Gemini 3 Flash (Preview) | 64.3 |
| 11 | Claude Opus 4.5 (20251101) | 63.1 |
| 12 | Claude Opus 4.5 (20251101) (Unknown) | 63.1 |
| 13 | GPT-5.1 Codex Mini | 61.6 |
| 14 | GPT-5.1 Codex Max | 60.4 |
| 15 | GPT-5.1 Codex | 57.8 |
Interactive version: theaggregate.ai/benchmark?slug=terminal-bench-2-0 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.