Atlas

Benchmarks

← All benchmarks

ArxivMath March-April 2026 - four-run mean

Math

71 research-level math questions from March (30) and April (41) 2026, evaluated without tools and averaged over four runs per problem. Shares the ArxivMath group; repeated runs are not independent task counts.

Top models (higher is better)

ModelScore
Claude Fable 578.5
Loading Atlas data…