benchgap
Calibration

MathVision → VideoMMMU

VideoMMMU is estimated from MathVision with a Hill curve fitted on 7 models measured on both: y = 0.8294 + (1.2000 − 0.8294)·x^6.00 / (1.31861^6.00 + x^6.00), R² = 0.34, cross-validated error 1.9 pp. It is used for 2 estimates.

Estimated modelMathVisionVideoMMMUSource
Qwen3.5-27B86.0%85.6%estimated ± 1.9 pp, low confidence
Qwen3.5-35B-A3B83.9%85.2%estimated ± 1.9 pp, low confidence