Terminal-Bench 4.0 — ML
Agents · 2026-09-22
Resolution rate on the 11 machine-learning tasks in Terminal-Bench 4.0.
Top models (higher is better)
| Model | Score |
|---|---|
| Kimi K3.1 Pro | 81.8 |
| Gemini 4 Carbon | 78.8 |
| Gemini 4 Argon | 69.7 |
| GPT-6 Astra | 66.7 |
| Muse Wonder Full | 66.7 |
| Claude Haiku 5.5 | 63.6 |
| GLM-5.3 | 54.5 |
| GPT-6.1 Sol | 54.5 |
| GLM-5.3 Flash | 51.5 |
| Qwen3.8 Max (0902) | 51.5 |
| GPT-6 Sol | 45.5 |
| Muse Wonder | 45.5 |
| DeepSeek V4.1 Flash 1017 | 42.4 |
| GPT-5.6 Sol | 42.4 |
| GPT-6.1 Terra | 42.4 |
| Muse Spark 1.3 | 42.4 |
| GPT-5.6 Terra | 36.4 |
| Grok 4.7 | 36.4 |
| MiniMax M3.1 Flash | 36.4 |
| Mistral Large 4 Preview | 33.3 |
| MiMo-V2.6-Pro | 30.3 |
| Step 5 Preview | 30.3 |
| Opus 4.8 | 27.3 |
| Ember-1 | 27.3 |
| DeepSeek V4.1 Flash | 24.2 |
| Grok 4.6 | 24.2 |
| Kimi K3 | 21.2 |
| MiMo-V2.6-Flash | 21.2 |
| Sonnet 5 | 18.2 |
| DeepSeek V4 Pro 0813 | 18.2 |
| Gemini 3.8 Flash | 18.2 |
| GPT-5.6 Luna | 18.2 |
| DeepSeek-V4-Flash-0731 | 15.2 |
| Gemini 3.7 Flash | 12.1 |
| Hy4 Preview | 12.1 |
| GPT-6 Luna | 9.1 |
| Gemini 3.5 Flash | 6.1 |
| Mercury 3 | 6.1 |
| Beam 1.0 501B A23B | 3.0 |
| Grok 4.5 | 3.0 |