Calibration
AA ITBench → AA Tau3 Banking
AA Tau3 Banking is estimated from AA ITBench with a linear curve fitted on 7 models measured on both: y = -0.2865·x + 0.6024, R² = 0.34, cross-validated error 3.0 pp. It is used for 10 estimates.
| Estimated model | AA ITBench | AA Tau3 Banking | Source |
|---|---|---|---|
| Claude Opus 4.7 (Adaptive) | 46.7% | 46.9% | estimated ± 3.0 pp, low confidence |
| Claude Opus 5.5 | 38.2% | 49.3% | estimated ± 3.0 pp, low confidence |
| DeepSeek V4.1 Flash | 46.9% | 46.8% | estimated ± 3.0 pp, low confidence |
| GLM-5.2 | 42.7% | 48.0% | estimated ± 3.0 pp, low confidence |
| GPT-5.5 | 45.8% | 47.1% | estimated ± 3.0 pp, low confidence |
| GPT-5.6 Sol | 56.2% | 44.1% | estimated ± 3.0 pp, low confidence |
| GPT-5.6 Terra | 51.0% | 45.6% | estimated ± 3.0 pp, low confidence |
| GPT-6 Sol | 49.4% | 46.1% | estimated ± 3.0 pp, low confidence |
| Grok 4.7 | 42.1% | 48.2% | estimated ± 3.0 pp, low confidence |
| Step 5 Preview | 55.6% | 44.3% | estimated ± 3.0 pp, low confidence |