MATH-500: leaderboard

500 problems sampled uniformly at random from the MATH competition-math test set (Hendrycks et al.) by OpenAI for Let's Verify Step by Step (2023); exact-match on the final answer.

Source: arxiv.org.

Interactive version: theaggregate.ai/benchmark?slug=math-500 · How It Works · Data refreshed daily, snapshot 2026-09-05.