Definition
Plain language
A U.S. high school math competition used as a benchmark for AI math reasoning.
As stated in the literature
The American Mathematics Competitions, a series of high school math contests whose problems are used as evaluation tasks in math-reasoning benchmark suites.
Why it matters: It provides a broad range of difficulty levels for testing how well AI handles competition-style mathematical reasoning.
For example, AMC problems range from approachable multiple-choice questions to harder ones that filter contestants toward national competitions.
Heard on the show
“Across six math benchmarks — MATH-500, GSM8K, AMC, AIME, Minerva, OlympiadBench — and across model families and sizes, ReasonMaxxer matches or exceeds full RL pipelines on the majority of comparisons.”Episode 026 — What RL Actually Does to Language Models, at the Token Level