Calibration
ExploitGym → OSWorld 2.0
OSWorld 2.0 is estimated from ExploitGym with a linear curve fitted on 7 models measured on both: y = 1.4613·x + 0.1463, R² = 0.75, cross-validated error 13.6 pp. It is used for 1 estimate.
| Estimated model | ExploitGym | OSWorld 2.0 | Source |
|---|---|---|---|
| Claude Mythos Preview | 17.5% | 40.2% | estimated ± 13.6 pp, low confidence |