MATH-500

MATH-500 is a representative 500-problem subset of the MATH dataset: 500 competition-level math problems from middle school up to Olympiad qualifying level. Each answer is graded exactly, so the score shows how reliably a model solves classic math problems without any tools.

MATH-500 measures mathematical problem solving. In detail, it checks:

• competition and school math from easy to olympiad qualifying level
• exact numeric/symbolic final answers
• correct multi-step derivations
• solving without calculators or code execution

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-500 benchmark?

MATH-500 is a representative 500-problem subset of the MATH dataset: 500 competition-level math problems from middle school up to Olympiad qualifying level. Each answer is graded exactly, so the score shows how reliably a model solves classic math problems without any tools.

What does MATH-500 test?

MATH-500 mainly measures mathematical problem solving. Concretely, it checks: competition and school math from easy to olympiad qualifying level; exact numeric/symbolic final answers; correct multi-step derivations.

How is the MATH-500 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-500 data and scores?

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-500/scores) or pull it directly from the Open LLM Radar API with any HTTP client.