benchgap
Calibration

ScreenSpot Pro → AA-MMMU-Pro

AA-MMMU-Pro is estimated from ScreenSpot Pro with a offset logistic curve fitted on 12 models measured on both: y = 0.5898 + (0.7960 − 0.5898) / (1 + exp(−200.00·(x − 0.6734))), R² = 0.71, cross-validated error 9.4 pp. It is used for 4 estimates.

Estimated modelScreenSpot ProAA-MMMU-ProSource
Holo2-235B-A22B70.6%79.6%estimated ± 9.4 pp, medium confidence
Holo2-30B-A3B66.1%60.6%estimated ± 9.4 pp, medium confidence
Holo2-4B57.2%59.0%estimated ± 9.4 pp, medium confidence
Holo2-8B58.9%59.0%estimated ± 9.4 pp, medium confidence