Terminal-Bench Hard
Agents · 2025-12-22
Terminal-Bench Hard is the difficult subset of a terminal agent benchmark covering software engineering, system administration, data processing, compilation, model training, debugging, and related shell workflows. Higher scores mean more terminal tasks completed.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 65.9 |
| Claude Fable 5 | 62.9 |
| GPT-5.6 Terra | 62.9 |
| GPT-5.5 | 60.6 |
| Opus 4.8 | 58.3 |
| GPT-5.4 | 57.6 |
| Opus 4.7 | 54.5 |
| Gemini 3.1 Pro Preview | 53.8 |
| Sonnet 4.6 | 53.0 |
| GPT-5.3-Codex | 53.0 |
| GPT-5.4 Mini | 52.3 |
| GLM-5.2 | 50.8 |
| Qwen3.7-Max | 50.8 |
| KAT-Coder-Pro V2 | 49.2 |
| Opus 4.6 | 48.5 |
| Opus 4.5 | 47.0 |
| GPT-5.2 | 47.0 |
| Qwen3.7-Plus | 47.0 |
| DeepSeek-V4-Pro | 46.2 |
| Gemini 3.5 Flash | 46.2 |
| GPT-5.1 | 45.5 |
| Muse Spark | 45.5 |
| Kimi K2.7 Code | 44.7 |
| Kimi K2.6 | 43.9 |
| Qwen3.6-Max-Preview | 43.9 |
| Qwen3.6 Plus (2026-04-02) | 43.9 |
| GLM-5.1 | 43.2 |
| GLM-5 | 43.2 |
| MiMo-V2.5-Pro | 43.2 |
| GPT-5.4 Nano | 42.4 |
| GPT-5.5 Instant | 42.4 |
| MiniMax M3 | 42.4 |
| Gemini 3 Pro Preview | 41.7 |
| MiMo-V2.5 | 41.7 |
| Grok 4.20 | 40.9 |
| Qwen3.5 397B A17B | 40.9 |
| MiMo-V2-Pro | 40.9 |
| MiniMax M2.7 | 39.4 |
| DeepSeek-V4-Flash | 38.6 |
| Gemini 3 Flash Preview | 38.6 |