benchgap
Calibration

SimpleVQA → RealWorldQA

RealWorldQA is estimated from SimpleVQA with a Michaelis–Menten curve fitted on 5 models measured on both: y = 1.0454·x / (0.14516 + x), R² = 0.95, cross-validated error 3.3 pp. It is used for 0 estimates.