Glossary · Term

FrontierMath

← all terms

Definition

Plain language

A benchmark of research-grade math problems used to test AI math reasoning.

As stated in the literature

An Epoch AI benchmark of advanced mathematics problems organized into difficulty tiers (Tier 4 being short-term research projects for PhD mathematicians), used to evaluate AI mathematical reasoning capabilities.

Also called: FrontierScience-Research

Why it matters: It pushes math benchmarks past competition-style problems into territory where solving anything is a meaningful capability signal.

For example, a Tier 4 FrontierMath problem might be the kind of question a math PhD student would spend a week or two working on as a self-contained research mini-project.

Heard on the show

“SU-01, despite never being trained on chemistry or biology, gets about twelve percent on something called FrontierScience-Research — which is a benchmark for general scientific reasoning.”
Episode 048 — How a 30B Open Model Reached Olympiad Gold With the Right Recipe

Mentioned in 2 episodes

  1. 048
    How a 30B Open Model Reached Olympiad Gold With the Right Recipe
  2. 029
    Why Forty-Eight Percent on FrontierMath Isn't the Real Story in DeepMind's New Math Paper

Related concepts