benchgap
Calibration

ScreenSpot Pro → VideoMMMU

VideoMMMU is estimated from ScreenSpot Pro with a Hill curve fitted on 6 models measured on both: y = 0.8391 + (1.2000 − 0.8391)·x^6.00 / (1.18486^6.00 + x^6.00), R² = 0.59, cross-validated error 1.7 pp. It is used for 5 estimates.

Estimated modelScreenSpot ProVideoMMMUSource
GPT-6 Astra92.7%90.6%estimated ± 1.7 pp, low confidence
Holo2-235B-A22B70.6%85.5%estimated ± 1.7 pp, medium confidence
Holo2-30B-A3B66.1%85.0%estimated ± 1.7 pp, medium confidence
Holo2-4B57.2%84.4%estimated ± 1.7 pp, medium confidence
Holo2-8B58.9%84.4%estimated ± 1.7 pp, medium confidence