Atlas

Benchmarks

← All benchmarks

ArXivLean 03/2026

Math · 2026-04-21

This MathArena track evaluates tool-assisted formal theorem proving on 41 Lean 4.29.0 statements formalized from March 2026 arXiv papers. Scores report the percentage of problems whose final Lean proof passes MathArena's checker, averaged across observed problem attempts. The item count records distinct problems, not a single-run binomial denominator.

Top models (higher is better)

ModelScore
Aristotle17.1
GPT-5.417.1
GPT-5.517.1
Leanstral 1.5 119B A6B17.1
Opus 4.714.6
Gemini 3.1 Pro Preview14.6
GLM-5.12.4
Step 3.5 Flash0.0
Loading Atlas data…