MIT researchers have compiled the largest collection of math Olympiad problems, aggregating data from 47 countries to create an extensive open dataset. This collection addresses limitations in existing benchmarks by enhancing size, language coverage, and task diversity, as detailed in the MIT Arxiv paper (arxiv.org/pdf/2604.18584). The dataset is positioned as a critical resource for training large language and multimodal AI models in mathematical problem solving. While the Art of Problem Solving (AoPS) forums contain a substantial problem archive, MIT's collection is officially recognized as the largest curated dataset.
AI researchers developing large language and multimodal models, educators and curriculum developers in advanced mathematics, data scientists curating training datasets, and competitive math coaches seeking comprehensive problem collections.
If I am not mistaken, AI companies would train their models on that dataset.jedishrfu said:TL;DR: Math Olympiad problems MIT collection
https://news.mit.edu/2026/mit-scien...ection-olympiad-level-math-problems-open-0424
MIT researchers have built the largest collection of math Olympiad problems from 47 countries.