Calibration
Terminal-Bench 4.1 → Terminal-Bench Hard
Terminal-Bench Hard is estimated from Terminal-Bench 4.1 with a linear curve fitted on 5 models measured on both: y = 0.2775·x + 0.5245, R² = 0.90, cross-validated error 2.8 pp. It is used for 41 estimates.
| Estimated model | Terminal-Bench 4.1 | Terminal-Bench Hard | Source |
|---|---|---|---|
| Claude Fable 5.2 | 65.8% | 70.7% | estimated ± 2.8 pp, low confidence |
| Claude Opus 5.5 | 65.8% | 70.7% | estimated ± 2.8 pp, low confidence |
| Gemini 4 Carbon | 64.6% | 70.4% | estimated ± 2.8 pp, low confidence |
| Claude Sonnet 5.5 | 63.9% | 70.2% | estimated ± 2.8 pp, low confidence |
| GPT-6.1 Sol | 60.9% | 69.4% | estimated ± 2.8 pp, low confidence |
| GPT-6 Astra | 60.3% | 69.2% | estimated ± 2.8 pp, low confidence |
| Kimi K3.1 Pro | 59.7% | 69.0% | estimated ± 2.8 pp, low confidence |
| Claude Fable 5.1 | 57.6% | 68.4% | estimated ± 2.8 pp, low confidence |
| Gemini 4 Hydrogen | 56.4% | 68.1% | estimated ± 2.8 pp, low confidence |
| Gemini 4 Argon | 55.5% | 67.9% | estimated ± 2.8 pp, low confidence |
| Claude Opus 5 | 53.3% | 67.2% | estimated ± 2.8 pp, low confidence |
| GPT-6.1 Terra | 53.3% | 67.2% | estimated ± 2.8 pp, low confidence |
| GPT-6 Sol | 50.3% | 66.4% | estimated ± 2.8 pp, low confidence |
| Muse Wonder | 47.0% | 65.5% | estimated ± 2.8 pp, low confidence |
| GLM-5.3 | 41.2% | 63.9% | estimated ± 2.8 pp, medium confidence |
| Claude Haiku 5.5 | 40.0% | 63.6% | estimated ± 2.8 pp, medium confidence |
| Grok 4.7 | 38.8% | 63.2% | estimated ± 2.8 pp, medium confidence |
| DeepSeek V4.1 Flash 1017 | 38.5% | 63.1% | estimated ± 2.8 pp, medium confidence |
| Hy4 | 37.9% | 63.0% | estimated ± 2.8 pp, medium confidence |
| GLM-5.3-Flash | 37.0% | 62.7% | estimated ± 2.8 pp, medium confidence |
| Muse Ember | 30.6% | 60.9% | estimated ± 2.8 pp, medium confidence |
| Mistral Large 4 | 27.6% | 60.1% | estimated ± 2.8 pp, medium confidence |
| MiMo-V2.6-Pro | 27.0% | 59.9% | estimated ± 2.8 pp, medium confidence |
| Kimi K3 | 23.0% | 58.8% | estimated ± 2.8 pp, medium confidence |
| Ember-1 | 21.8% | 58.5% | estimated ± 2.8 pp, medium confidence |
| GPT-5.6 Terra | 21.5% | 58.4% | estimated ± 2.8 pp, medium confidence |
| Grok 4.6 | 19.7% | 57.9% | estimated ± 2.8 pp, medium confidence |
| Gemini 3.8 Flash | 18.2% | 57.5% | estimated ± 2.8 pp, medium confidence |
| GPT-5.6 Luna | 17.6% | 57.3% | estimated ± 2.8 pp, medium confidence |
| MiMo-V2.6-Flash | 16.1% | 56.9% | estimated ± 2.8 pp, medium confidence |
| GPT-6 Luna | 15.2% | 56.7% | estimated ± 2.8 pp, medium confidence |
| Muse Spark 1.3 | 14.9% | 56.6% | estimated ± 2.8 pp, medium confidence |
| Gemini 3.7 Flash | 12.4% | 55.9% | estimated ± 2.8 pp, medium confidence |
| Grok 4.5 | 12.4% | 55.9% | estimated ± 2.8 pp, medium confidence |
| Claude Sonnet 5 | 12.1% | 55.8% | estimated ± 2.8 pp, medium confidence |
| MiniMax M3 | 7.6% | 54.6% | estimated ± 2.8 pp, medium confidence |
| DeepSeek V4 Pro | 6.7% | 54.3% | estimated ± 2.8 pp, medium confidence |
| Beam 1.0 501B-A23B | 6.4% | 54.2% | estimated ± 2.8 pp, medium confidence |
| Inkling-Small | 0.9% | 52.7% | estimated ± 2.8 pp, low confidence |
| Nemotron 3.5 Super VL 120B A12B | 0.3% | 52.5% | estimated ± 2.8 pp, low confidence |
| Inkling | 0.0% | 52.5% | estimated ± 2.8 pp, low confidence |
Fitted on
| Model | Terminal-Bench 4.1 | Terminal-Bench Hard |
|---|---|---|
| Claude Fable 5 | 44.9% | 62.9% |
| Claude Opus 4.8 | 23.6% | 58.3% |
| Gemini 3.1 Pro | 2.7% | 53.8% |
| GPT-5.4 mini | 1.8% | 52.3% |
| GPT-5.6 Sol | 38.5% | 65.9% |