Terminal-Bench 4.0 (Artificial Analysis)
Agents
Artificial Analysis' implementation of all 66 Terminal-Bench 4.0 tasks, reporting pass@1 averaged over three repeats per task. Version 4.0 changes task membership, compute and time allowances, instructions and verification; it is separate from Terminal-Bench 2.1.
Top models (higher is better)
| Model | Score |
|---|---|
| Kimi K3.1 Pro | 67.2 |
| Gemini 4 Carbon | 65.2 |
| Claude Sonnet 5.5 | 63.6 |
| Claude Opus 5.5 | 59.6 |
| GPT-6 Astra | 59.6 |
| Gemini 4 Argon | 57.1 |
| Muse Wonder Full | 57.1 |
| GPT-6.1 Sol | 56.1 |
| GPT-6.1 Terra | 55.6 |
| Claude Fable 5.1 | 55.1 |
| Muse Wonder | 52.0 |
| Claude Opus 5 | 49.0 |
| GPT-6 Sol | 43.9 |
| Claude Fable 5 | 42.4 |
| GLM-5.3 | 41.9 |
| GPT-5.6 Sol | 39.9 |
| Qwen3.8 Max (0902) | 38.9 |
| GPT-5.6 Terra | 35.4 |
| MiMo-V2.6-Pro | 34.8 |
| DeepSeek V4.1 Flash 1017 | 33.8 |
| Ling 3.1 Flash | 33.3 |
| Muse Spark 1.3 | 33.3 |
| Step 5 Preview | 33.3 |
| Claude Haiku 5.5 | 32.8 |
| GLM-5.3 Flash | 32.8 |
| Ling 3.1 Flash 1019 | 32.8 |
| DeepSeek V4.1 Flash | 26.8 |
| Mistral Large 4 Preview | 26.8 |
| Grok 4.7 | 25.8 |
| MiniMax M3.1 Flash | 25.8 |
| Qwen3.8-Flash-Next | 25.3 |
| MiMo-V2.6-Flash | 22.7 |
| Opus 4.8 | 21.7 |
| Grok 4.6 | 21.2 |
| Gemini 3.8 Flash | 19.7 |
| Qwen3.8-Max | 18.7 |
| DeepSeek-V4-Pro | 14.6 |
| GPT-5.5 | 14.6 |
| Sonnet 5 | 14.1 |
| DeepSeek V4 Pro 0813 | 14.1 |