MATH: 10,000+ Competition Math Problems with Step-by-Step Solutions

The Mathematics Aptitude Test of Heuristics (MATH) dataset contains between 10,000 and 100,000 competition-level math problems from sources like AMC 10, AMC 12, and AIME. Released by qwedsacf and last updated in 2023, it provides expert-generated step-by-step solutions for every problem.

Use Cases

• Training models for explanation-generation using the step-by-step solution text
• Benchmarking mathematical reasoning on AMC 10 and AIME level competition problems
• Fine-tuning language models for chain-of-thought reasoning in technical domains

Strengths

• 10,000 to 100,000 records
• Expert-generated step-by-step solutions
• MIT license for open redistribution

Limitations

• Monolingual English focus
• High difficulty ceiling may not represent general-purpose mathematical knowledge

Provenance

Source

Mathematics Aptitude Test of Heuristics (MATH) / Arxiv 2103.03874

Collection Method

Expert Generated

Freshness

Last updated January 2023