Calibration
OSWorld-Verified → AA Agentic Index
AA Agentic Index is estimated from OSWorld-Verified with a inverse Michaelis–Menten curve fitted on 17 models measured on both: y = 0.08640·(x − 0.0000) / (0.0000 + 1.0050 − x), R² = 0.70, cross-validated error 8.0 pp. It is used for 4 estimates.
| Estimated model | OSWorld-Verified | AA Agentic Index | Source |
|---|---|---|---|
| Holo3-122B-A10B | 78.9% | 31.5% | estimated ± 8.0 pp, medium confidence |
| Holo3-35B-A3B | 82.6% | 39.8% | estimated ± 8.0 pp, medium confidence |
| UI-Mate-27B | 77.0% | 28.3% | estimated ± 8.0 pp, medium confidence |
| UI-Mate-9B | 66.2% | 16.7% | estimated ± 8.0 pp, medium confidence |