benchgap
Calibration

ZeroBench → BabyVision

BabyVision is estimated from ZeroBench with a Michaelis–Menten curve fitted on 5 models measured on both: y = 0.8899·x / (0.05528 + x), R² = 0.30, cross-validated error 13.1 pp. It is used for 0 estimates.

Fitted on

ModelZeroBenchBabyVision
dots3-note Preview19.0%50.0%
Muse Spark Open50.0%80.2%
Qwen3.8 Max24.0%82.0%
Seed 2.1 Pro18.0%73.7%
Seed 2.1 Turbo11.0%62.9%