Calibration
Terminal-Bench 2.1 (Vals AI) → AA Terminal-Bench 4.0
AA Terminal-Bench 4.0 is estimated from Terminal-Bench 2.1 (Vals AI) with a linear curve fitted on 7 models measured on both: y = 1.3076·x + -0.6494, R² = 0.72, cross-validated error 14.0 pp. It is used for 29 estimates.
| Estimated model | Terminal-Bench 2.1 (Vals AI) | AA Terminal-Bench 4.0 | Source |
|---|---|---|---|
| Claude Opus 5 | 84.6% | 45.7% | estimated ± 14.0 pp, low confidence |
| Grok 4.6 | 78.3% | 37.4% | estimated ± 14.0 pp, low confidence |
| GPT-5.5 | 76.4% | 35.0% | estimated ± 14.0 pp, low confidence |
| Gemini 3.6 Flash | 73.8% | 31.6% | estimated ± 14.0 pp, low confidence |
| Gemini 3.1 Pro | 70.8% | 27.6% | estimated ± 14.0 pp, low confidence |
| Claude Opus 4.7 | 68.5% | 24.6% | estimated ± 14.0 pp, low confidence |
| Kimi K2.7 Code | 67.0% | 22.7% | estimated ± 14.0 pp, low confidence |
| Qwen3.7 Max | 61.0% | 14.8% | estimated ± 14.0 pp, low confidence |
| MiMo-V2.5 | 60.7% | 14.4% | estimated ± 14.0 pp, low confidence |
| Composer 2.5 | 58.4% | 11.4% | estimated ± 14.0 pp, low confidence |
| Claude Sonnet 4.6 | 57.3% | 10.0% | estimated ± 14.0 pp, low confidence |
| MiMo-V2.5-Pro | 57.3% | 10.0% | estimated ± 14.0 pp, low confidence |
| GLM-5.1 | 56.9% | 9.5% | estimated ± 14.0 pp, low confidence |
| GPT-5.4 mini | 54.7% | 6.6% | estimated ± 14.0 pp, low confidence |
| Gemini 3 Flash | 53.9% | 5.5% | estimated ± 14.0 pp, low confidence |
| Kimi K2.6 | 53.6% | 5.1% | estimated ± 14.0 pp, low confidence |
| Qwen3.6 Plus | 53.2% | 4.6% | estimated ± 14.0 pp, low confidence |
| Qwen3.7 Plus | 52.8% | 4.1% | estimated ± 14.0 pp, low confidence |
| Claude Haiku 4.5 | 43.8% | 0.0% | estimated ± 14.0 pp, low confidence |
| Command A+ | 17.6% | 0.0% | estimated ± 14.0 pp, low confidence |
| Gemini 3.1 Flash-Lite | 34.1% | 0.0% | estimated ± 14.0 pp, low confidence |
| GPT-5.4 nano | 41.6% | 0.0% | estimated ± 14.0 pp, low confidence |
| Grok 4.20 | 44.2% | 0.0% | estimated ± 14.0 pp, low confidence |
| Grok 4.3 | 41.9% | 0.0% | estimated ± 14.0 pp, low confidence |
| Laguna M.1 | 34.1% | 0.0% | estimated ± 14.0 pp, low confidence |
| Laguna XS.2 | 25.8% | 0.0% | estimated ± 14.0 pp, low confidence |
| Mercury 2.5 | 34.5% | 0.0% | estimated ± 14.0 pp, low confidence |
| MiniMax M2.7 | 48.7% | 0.0% | estimated ± 14.0 pp, low confidence |
| Mistral Medium 3.5 128B | 39.0% | 0.0% | estimated ± 14.0 pp, low confidence |
Fitted on
| Model | Terminal-Bench 2.1 (Vals AI) | AA Terminal-Bench 4.0 |
|---|---|---|
| Claude Fable 5.1 | 85.0% | 52.0% |
| Gemini 3.8 Flash | 81.3% | 19.7% |
| GLM-5.3 | 71.5% | 41.9% |
| GPT-6 Astra | 87.3% | 59.1% |
| Grok 4.7 | 73.4% | 25.8% |
| Inkling | 47.6% | 1.0% |
| Qwen3.8-27B | 58.4% | 5.6% |