Benchmarks

MiniF2F (test)

Mathunit % · normalized over [0, 94]

Formal (Lean 4) theorem-proving benchmark, test split.

#ModelSourceScoreNormalized

Score distribution

2 tracked results across the normalization window

094

Score vs. parameters

Open-weights models, log-x params

Only one open-weights model with a disclosed parameter count — not enough to plot a trend.