MIT scientists launch MathNet, creating largest Olympiad math problem collection

8 articles · Updated · MIT News · Apr 24

MathNet compiles over 30,000 expert-authored problems and solutions from 47 countries, 17 languages, and 143 competitions, making it five times larger than previous datasets.
The dataset, built with contributions from MIT, KAUST, and HUMAIN, offers high-quality, peer-reviewed problems and solutions, providing a valuable resource for both AI researchers and students worldwide.
MathNet highlights AI's uneven performance across languages and problem types, and introduces new benchmarks for mathematical reasoning, aiming to broaden both human and AI understanding of global mathematical traditions.

Can a massive math dataset truly level the playing field for aspiring mathematicians globally?

If AI can't be trusted with math proofs, how can we trust it with our critical software?

Beyond solving old problems, when will an AI generate truly novel mathematical ideas?

Does mastering Olympiad problems measure an AI's reasoning or just its test-taking skills?

Why do top AIs like GPT-5 still fail at one-in-three Olympiad math problems?

How will Saudi Arabia's massive AI investment, via HUMAIN, shape the future of machine reasoning?