Calibration
DeepSearchQA → OSWorld-Verified
OSWorld-Verified is estimated from DeepSearchQA with a Michaelis–Menten curve fitted on 6 models measured on both: y = 1.3601·x / (0.66062 + x), R² = 0.37, cross-validated error 6.7 pp. It is used for 2 estimates.
| Estimated model | DeepSearchQA | OSWorld-Verified | Source |
|---|---|---|---|
| Apodex 1.1 | 92.4% | 79.3% | estimated ± 6.7 pp, low confidence |
| Mercury 2.5 | 34.0% | 46.2% | estimated ± 6.7 pp, low confidence |