Calibration
AutomationBench → τ³-Bench Banking
τ³-Bench Banking is estimated from AutomationBench with a offset logistic curve fitted on 15 models measured on both: y = 0.2052 + (0.4740 − 0.2052) / (1 + exp(−153.25·(x − 0.3718))), R² = 0.90, cross-validated error 5.0 pp. It is used for 8 estimates.
| Estimated model | AutomationBench | τ³-Bench Banking | Source |
|---|---|---|---|
| GPT-6 Astra (high) | 66.6% | 47.4% | estimated ± 5.0 pp, medium confidence |
| GPT-6 Astra (xhigh) | 67.2% | 47.4% | estimated ± 5.0 pp, medium confidence |
| Gemini 4 Argon (high) | 77.5% | 47.4% | estimated ± 5.0 pp, low confidence |
| MiMo-V2.6-Pro | 58.6% | 47.4% | estimated ± 5.0 pp, medium confidence |
| Mistral Large 4 Preview | 59.9% | 47.4% | estimated ± 5.0 pp, medium confidence |
| GPT-6 Luna (max) | 53.2% | 47.4% | estimated ± 5.0 pp, medium confidence |
| GPT-6.1 Sol (xhigh) | 66.6% | 47.4% | estimated ± 5.0 pp, medium confidence |
| Grok 4.6 (xhigh) | 67.0% | 47.4% | estimated ± 5.0 pp, medium confidence |