FunctionalMATH: leaderboard

MATH problems rewritten by Consequent AI (2024) as parameterised templates re-instantiated in monthly snapshots; scored by the static-to-functional accuracy drop (58.35% to 80.31% at release).

Source: arxiv.org.

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