Calibration
Terminal-Bench 2.1 (Vals AI) → Terminal-Bench 2.1
Terminal-Bench 2.1 is estimated from Terminal-Bench 2.1 (Vals AI) with a Hill curve fitted on 29 models measured on both: y = 0.0000 + (0.8870 − 0.0000)·x^6.00 / (0.43383^6.00 + x^6.00), R² = 0.67, cross-validated error 6.9 pp. It is used for 26 estimates.
| Estimated model | Terminal-Bench 2.1 (Vals AI) | Terminal-Bench 2.1 | Source |
|---|---|---|---|
| Gemini 3.6 Flash | 73.8% | 85.2% | estimated ± 6.9 pp, medium confidence |
| Gemini 3.1 Pro | 70.8% | 84.2% | estimated ± 6.9 pp, medium confidence |
| Claude Opus 4.7 | 68.5% | 83.3% | estimated ± 6.9 pp, medium confidence |
| Kimi K2.7 Code | 67.0% | 82.6% | estimated ± 6.9 pp, medium confidence |
| Qwen3.7 Max | 61.0% | 78.5% | estimated ± 6.9 pp, medium confidence |
| MiMo-V2.5 | 60.7% | 78.3% | estimated ± 6.9 pp, medium confidence |
| Composer 2.5 | 58.4% | 75.9% | estimated ± 6.9 pp, medium confidence |
| Claude Sonnet 4.6 | 57.3% | 74.6% | estimated ± 6.9 pp, medium confidence |
| MiMo-V2.5-Pro | 57.3% | 74.6% | estimated ± 6.9 pp, medium confidence |
| GLM-5.1 | 56.9% | 74.1% | estimated ± 6.9 pp, medium confidence |
| GPT-5.4 mini | 54.7% | 71.0% | estimated ± 6.9 pp, medium confidence |
| Gemini 3 Flash | 53.9% | 69.7% | estimated ± 6.9 pp, medium confidence |
| Kimi K2.6 | 53.6% | 69.2% | estimated ± 6.9 pp, medium confidence |
| Qwen3.6 Plus | 53.2% | 68.5% | estimated ± 6.9 pp, medium confidence |
| Qwen3.7 Plus | 52.8% | 67.8% | estimated ± 6.9 pp, medium confidence |
| MiniMax M2.7 | 48.7% | 59.1% | estimated ± 6.9 pp, medium confidence |
| Grok 4.20 | 44.2% | 46.8% | estimated ± 6.9 pp, low confidence |
| Claude Haiku 4.5 | 43.8% | 45.6% | estimated ± 6.9 pp, low confidence |
| Grok 4.3 | 41.9% | 39.7% | estimated ± 6.9 pp, low confidence |
| GPT-5.4 nano | 41.6% | 38.8% | estimated ± 6.9 pp, low confidence |
| Mistral Medium 3.5 128B | 39.0% | 30.6% | estimated ± 6.9 pp, low confidence |
| Mercury 2.5 | 34.5% | 17.9% | estimated ± 6.9 pp, low confidence |
| Gemini 3.1 Flash-Lite | 34.1% | 16.9% | estimated ± 6.9 pp, low confidence |
| Laguna M.1 | 34.1% | 16.9% | estimated ± 6.9 pp, low confidence |
| Laguna XS.2 | 25.8% | 3.8% | estimated ± 6.9 pp, low confidence |
| Command A+ | 17.6% | 0.4% | estimated ± 6.9 pp, low confidence |
Fitted on
| Model | Terminal-Bench 2.1 (Vals AI) | Terminal-Bench 2.1 |
|---|---|---|
| Claude Fable 5 | 80.5% | 84.3% |
| Claude Opus 4.8 | 71.9% | 74.6% |
| Claude Sonnet 5 | 74.5% | 80.4% |
| DeepSeek V4.1 Flash 1017 | 74.5% | 90.6% |
| DeepSeek V4 Flash 0731 | 67.0% | 82.7% |
| DeepSeek V4 Pro 0813 | 54.7% | 87.9% |
| Gemini 3.5 Flash | 74.2% | 76.2% |
| Gemini 3.5 Flash-Lite | 50.2% | 54.0% |
| Gemini 3.7 Flash | 77.5% | 85.8% |
| Gemini 3.8 Flash | 81.3% | 89.4% |
| GLM-5.2 | 67.8% | 81.0% |
| GLM-5.3 | 71.5% | 88.2% |
| GLM-5.3-Flash | 62.9% | 84.3% |
| GPT-5.6 Luna | 79.0% | 84.7% |
| GPT-5.6 Sol | 85.8% | 91.9% |
| GPT-5.6 Terra | 77.5% | 87.4% |
| Grok 4.5 | 67.8% | 83.3% |
| Hy4 preview | 55.1% | 85.4% |
| Inkling | 47.6% | 63.8% |
| Inkling-Small | 55.1% | 64.7% |
| Kimi K3 | 80.9% | 88.3% |
| Ling 3.0 Flash | 50.2% | 57.0% |
| MiniMax M3 | 53.6% | 66.0% |
| Muse Spark 1.1 | 69.3% | 80.0% |
| Muse Spark 1.2 | 69.7% | 82.9% |
| Muse Spark 1.3 | 72.3% | 88.8% |
| Nemotron 3 Ultra | 50.9% | 56.4% |
| Qwen3.8-27B | 58.4% | 73.0% |
| Qwen3.8 Max | 67.4% | 86.6% |