Calibration
ScreenSpot Pro → MMMU-Pro
MMMU-Pro is estimated from ScreenSpot Pro with a Michaelis–Menten curve fitted on 11 models measured on both: y = 0.9612·x / (0.16079 + x), R² = 0.58, cross-validated error 2.7 pp. It is used for 5 estimates.
| Estimated model | ScreenSpot Pro | MMMU-Pro | Source |
|---|---|---|---|
| GPT-6 Astra | 92.7% | 81.9% | estimated ± 2.7 pp, medium confidence |
| Holo2-235B-A22B | 70.6% | 78.3% | estimated ± 2.7 pp, high confidence |
| Holo2-30B-A3B | 66.1% | 77.3% | estimated ± 2.7 pp, high confidence |
| Holo2-4B | 57.2% | 75.0% | estimated ± 2.7 pp, high confidence |
| Holo2-8B | 58.9% | 75.5% | estimated ± 2.7 pp, high confidence |