Terminal-Bench 4.0
Agents · 2026-08-28
Terminal-Bench 4.0 measures agents completing terminal tasks in Harbor. This major revision removes eight tasks, fixes nineteen tasks, calibrates CPU and memory resources, and gives each task an eight-hour agent timeout. Scores are mean resolution rates across repeated trials; they are not directly comparable to earlier Terminal-Bench task sets.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 66.4 |
| Kimi K3.1 Pro | 65.7 |
| Gemini 4 Carbon | 63.1 |
| Claude Sonnet 5.5 | 61.8 |
| GPT-6.1 Sol | 59.6 |
| GPT-6 Astra | 59.6 |
| Muse Wonder Full | 58.1 |
| Claude Fable 5.1 | 57.9 |
| Gemini 4 Argon | 57.6 |
| Claude Opus 5 | 53.9 |
| GPT-6.1 Terra | 50.3 |
| Muse Wonder | 49.5 |
| GPT-6 Sol | 49.4 |
| Claude Fable 5 | 44.5 |
| DeepSeek V4.1 Flash 1017 | 42.3 |
| GLM-5.3 | 41.8 |
| Claude Haiku 5.5 | 39.2 |
| Qwen3.8 Max (0902) | 38.9 |
| MiniMax M3.1 Flash | 38.1 |
| GPT-5.6 Sol | 37.9 |
| Grok 4.7 | 37.6 |
| GLM-5.3 Flash | 37.4 |
| MiMo-V2.6-Pro | 34.9 |
| GPT-5.6 Sol Pro | 32.3 |
| Step 5 Preview | 31.8 |
| DeepSeek V4.1 Flash | 31.2 |
| MiMo-V2.6-Flash | 28.8 |
| Muse Spark 1.3 | 27.8 |
| Mistral Large 4 Preview | 27.8 |
| Kimi K3 | 26.3 |
| GPT-5.6 Terra | 26.3 |
| Grok 4.6 | 25.3 |
| Nanbeige4.3-33B | 23.8 |
| Opus 4.8 | 23.6 |
| Ember-1 | 20.7 |
| Gemini 3.8 Flash | 19.7 |
| DeepSeek-V4-Flash-0731 | 18.7 |
| GPT-5.6 Luna | 17.3 |
| GPT-6 Luna | 16.4 |
| DeepSeek V4 Pro 0813 | 14.1 |