Calibration
OSWorld 2.0 → GDP.pdf
GDP.pdf is estimated from OSWorld 2.0 with a Michaelis–Menten curve fitted on 8 models measured on both: y = 0.2918·x / (0.08645 + x), R² = 0.76, cross-validated error 4.9 pp. It is used for 5 estimates.
| Estimated model | OSWorld 2.0 | GDP.pdf | Source |
|---|---|---|---|
| Claude Opus 4.7 (Adaptive) | 18.2% | 19.8% | estimated ± 4.9 pp, high confidence |
| Claude Opus 4.8 | 20.6% | 20.6% | estimated ± 4.9 pp, high confidence |
| Kimi K2.6 | 4.6% | 10.1% | estimated ± 4.9 pp, high confidence |
| Muse Spark 1.1 | 14.2% | 18.1% | estimated ± 4.9 pp, high confidence |
| Qwen3.7 Plus | 2.8% | 7.1% | estimated ± 4.9 pp, medium confidence |