Calibration
OSWorld-Verified → MCP Atlas
MCP Atlas is estimated from OSWorld-Verified with a Hill curve fitted on 13 models measured on both: y = 0.5000 + (1.2000 − 0.5000)·x^6.00 / (0.85528^6.00 + x^6.00), R² = 0.46, cross-validated error 11.5 pp. It is used for 4 estimates.
| Estimated model | OSWorld-Verified | MCP Atlas | Source |
|---|---|---|---|
| Holo3-122B-A10B | 78.9% | 76.6% | estimated ± 11.5 pp, low confidence |
| Holo3-35B-A3B | 82.6% | 81.3% | estimated ± 11.5 pp, low confidence |
| UI-Mate-27B | 77.0% | 74.3% | estimated ± 11.5 pp, low confidence |
| UI-Mate-9B | 66.2% | 62.4% | estimated ± 11.5 pp, low confidence |