AI Research
Apr 24, 2026
MIT and Partners Launch Comprehensive Dataset of Olympiad-Level Math Problems
Apr 24, 2026
AI Summary
Researchers from MIT, KAUST, and HUMAIN have created MathNet, the largest dataset of Olympiad-level math problems, featuring over 30,000 expert-authored problems from 47 countries. This resource aims to support students preparing for competitions and enhance AI models' mathematical reasoning capabilities.
- MathNet is a newly developed dataset containing more than 30,000 proof-based math problems and solutions, making it the largest of its kind.
- The dataset includes contributions from 47 countries and spans 17 languages, covering four decades of competition mathematics.
- It was created by researchers at MIT's CSAIL, King Abdullah University of Science and Technology, and the company HUMAIN, with significant contributions from community member Navid Safaei.
- MathNet draws exclusively from official national competition booklets, ensuring that the problems and solutions are expert-written and peer-reviewed.
- The dataset is designed to help students training for the International Mathematical Olympiad (IMO) and other competitions by providing a centralized, searchable collection of high-quality problems.
- Researchers aim to improve AI models' performance in mathematical reasoning using this dataset, which also serves as a benchmark for evaluating AI capabilities.
- Initial tests show that even top-performing AI models struggle with certain problems, particularly those involving visual reasoning or less common languages.
- The dataset includes benchmarks for recognizing structurally similar problems, which is important for both AI development and the math community.
- MathNet is publicly accessible, providing a valuable resource for students and educators worldwide.
datasetmath problemsai trainingeducationresearch