MATH

The MATH benchmark is a classic: 12,500 competition math problems collected from real competitions (AMC, AIME etc.), graded over five difficulty levels. It was designed specifically to expose weaknesses in multi-step mathematical thinking.

MATH measures mathematical problem solving. In detail, it checks:

• word and proof-problem math at five difficulty levels
• exact grading of final answers
• genuine multi-step reasoning
• comparability over years, as a widely used baseline

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

What is the MATH benchmark?

The MATH benchmark is a classic: 12,500 competition math problems collected from real competitions (AMC, AIME etc.), graded over five difficulty levels. It was designed specifically to expose weaknesses in multi-step mathematical thinking.

What does MATH test?

MATH mainly measures mathematical problem solving. Concretely, it checks: word and proof-problem math at five difficulty levels; exact grading of final answers; genuine multi-step reasoning.

How is the MATH score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get MATH data and scores?

The official source is https://arxiv.org/abs/2103.03874. The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/math/scores) or pull it directly from the Open LLM Radar API with any HTTP client.