benchgap
Calibration

BabyVision w/ Python → CharXiv w/o tools

CharXiv w/o tools is estimated from BabyVision w/ Python with a Michaelis–Menten curve fitted on 5 models measured on both: y = 1.4777·x / (0.64269 + x), R² = 0.62, cross-validated error 1.9 pp. It is used for 2 estimates.

Estimated modelBabyVision w/ PythonCharXiv w/o toolsSource
Glimpse95.4%88.3%estimated ± 1.9 pp, medium confidence
MiniMax M3.1 Flash91.0%86.6%estimated ± 1.9 pp, medium confidence

Fitted on

ModelBabyVision w/ PythonCharXiv w/o tools
Inkling 2 Small93.7%85.6%
Kimi K3.1 Pro95.4%89.0%
Kimi K385.7%84.8%
Qwen3.8-27B85.6%83.7%
Qwen3.8 Max91.3%88.4%