Dataset Download And Curation From Competition Sources

12.5K competition math problems — AMC/AIME/Olympiad level, 7 subjects, standard math benchmark.

Unique: Curates problems exclusively from official mathematical competitions (AMC, AIME, Olympiads) rather than synthetic or crowd-sourced problems, ensuring high quality and genuine mathematical reasoning requirements. Manual curation and verification provide confidence in problem correctness and difficulty calibration.

vs others: More rigorous than generic math QA datasets because problems are sourced from official competitions with established difficulty standards, making MATH the de facto benchmark for evaluating LLM mathematical reasoning in research.