Google and OpenAI Achieve Gold at International Math Olympiad with AI Models

Google’s DeepMind and OpenAI have made history, as their AI models earned gold medals at the 66th International Mathematical Olympiad (IMO), the first time general-purpose artificial intelligence has crossed the top-scoring threshold in the global high-school competition. Each model solved five out of six problems using natural language reasoning rather than traditional formal methods.

This achievement marks a significant leap in AI’s mathematical reasoning capabilities. OpenAI’s success came from scaling “test-time compute,” allowing models to process problems longer and run multiple reasoning paths in parallel. While expensive, this approach proved highly effective, according to OpenAI researcher Noam Brown.

Google’s DeepMind used a version of its Gemini Deep Think model—unveiled earlier this year—that solved problems within the official 4.5-hour time limit using only natural language. Unlike earlier AI attempts, this model didn’t rely on symbolic computation, showcasing the evolution of AI reasoning.

Junehyuk Jung, a Brown University professor and DeepMind researcher, suggested AI could soon assist in solving unsolved mathematical problems, highlighting a new era of AI-human collaboration. Jung, a former IMO gold medalist himself, noted the breakthrough also holds promise for fields beyond mathematics, such as physics.

Of 630 student competitors in Queensland, Australia, only 67 earned gold. IMO officials officially collaborated with AI companies for the first time, certifying and coordinating the public release of results. OpenAI posted its outcome on Saturday with board permission, while Google waited for IMO’s Monday approval.

These results demonstrate AI’s growing potential not just in academia, but in solving real-world problems through advanced reasoning, marking a milestone in the race toward artificial general intelligence.