AA Terminal-Bench Hard — leaderboard
Artificial Analysis independent evaluation of Terminal-Bench Hard: agentic terminal tasks including software engineering and system administration (44 tasks).
Metric: Accuracy (%). Source: artificialanalysis.ai. Status: saturation imminent. 432 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 65.91 |
| 2 | GPT-5.6 Terra (xHigh) | 62.88 |
| 3 | GPT-5.6 Sol (Medium) | 62.88 |
| 4 | GPT-5.6 Sol (High) | 62.12 |
| 5 | GPT-5.6 Sol (xHigh) | 61.36 |
| 6 | GPT-5.5 (xHigh) | 60.61 |
| 7 | GPT-5.6 Sol (Low) | 60.61 |
| 8 | GPT-5.5 (High) | 59.85 |
| 9 | Claude Opus 4.8 (Adaptive Reasoning, Max Effort) | 58.33 |
| 10 | GPT-5.4 (xHigh) | 57.58 |
| 11 | GPT-5.5 (Medium) | 57.58 |
| 12 | GPT-5.6 Terra (Max) | 57.58 |
| 13 | GPT-5.6 Terra (High) | 57.58 |
| 14 | Gemini 3.1 Pro (Preview) | 53.79 |
| 15 | Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort) | 53.03 |
Interactive version: theaggregate.ai/benchmark?slug=aa-terminal-bench-hard · How the rankings work · Data refreshed daily, snapshot 2026-07-22.