Calibration
OSWorld 2.0 → BrowseComp
BrowseComp is estimated from OSWorld 2.0 with a Hill curve fitted on 10 models measured on both: y = 0.8278 + (0.9364 − 0.8278)·x^6.00 / (0.54919^6.00 + x^6.00), R² = 0.83, cross-validated error 2.2 pp. It is used for 11 estimates.
| Estimated model | OSWorld 2.0 | BrowseComp | Source |
|---|---|---|---|
| Claude Fable 5.1 | 41.7% | 84.5% | estimated ± 2.2 pp, high confidence |
| Claude Opus 4.7 | 13.9% | 82.8% | estimated ± 2.2 pp, high confidence |
| Claude Opus 5.5 | 48.7% | 86.3% | estimated ± 2.2 pp, high confidence |
| Claude Sonnet 4.6 | 8.3% | 82.8% | estimated ± 2.2 pp, high confidence |
| Gemini 3.7 Flash | 47.9% | 86.1% | estimated ± 2.2 pp, high confidence |
| Gemini 3.8 Flash | 59.0% | 89.4% | estimated ± 2.2 pp, high confidence |
| Gemini 4 Argon | 69.2% | 91.5% | estimated ± 2.2 pp, high confidence |
| Muse Spark 1.3 | 66.9% | 91.1% | estimated ± 2.2 pp, high confidence |
| Qwen3.7 Plus | 2.8% | 82.8% | estimated ± 2.2 pp, medium confidence |
| Qwen3.8-Flash-Next | 19.4% | 82.8% | estimated ± 2.2 pp, high confidence |
| Qwen3.8 Max | 19.4% | 82.8% | estimated ± 2.2 pp, high confidence |