Calibration
JobBench → OSWorld-Verified
OSWorld-Verified is estimated from JobBench with a Michaelis–Menten + offset curve fitted on 14 models measured on both: y = 0.0000 + 1.1859·x / (0.24949 + x), R² = 0.44, cross-validated error 8.7 pp. It is used for 4 estimates.
| Estimated model | JobBench | OSWorld-Verified | Source |
|---|---|---|---|
| Claude 4.1 Opus | 21.9% | 55.4% | estimated ± 8.7 pp, low confidence |
| Claude Haiku 4.5 | 16.0% | 46.3% | estimated ± 8.7 pp, low confidence |
| GPT-5 (high) | 8.5% | 30.2% | estimated ± 8.7 pp, low confidence |
| Qwen3.5 Plus | 18.5% | 50.5% | estimated ± 8.7 pp, low confidence |