PolyMath Mathematical Reasoning Dataset

*This dataset supports online use.Click here to jump.

PolyMath is a mathematical reasoning dataset jointly released by Alibaba and Shanghai Jiao Tong University in 2025. The related paper results are:PolyMath: Evaluating Mathematical Reasoning in Multilingual Contexts", which aims to promote research in mathematical reasoning.

This dataset contains 500 high-quality mathematical reasoning questions, with 125 questions for each language level. It uses two core dimensions, depth of thinking and breadth of knowledge, to define and categorize mathematical difficulty across four levels, from K-12 to Olympiad competitions and advanced frontier mathematics.

Dataset features:

• Language Diversity: Every problem in PolyMath is available in 18 parallel languages, covering more than 75% native speakers and major language families worldwide, ensuring diversity in both high-resource and low-resource languages.

• High-quality annotation: Each question translation is calibrated by a linguist, avoiding the direct use of LLM-generated output to ensure accurate and clear terminology and logic.