Humans Thrash AI at 2025 International Mathematical Olympiad

In a decisive show of mathematical prowess, human contestants outperformed cutting-edge artificial intelligence models from Google and OpenAI at the 2025 International Mathematical Olympiad (IMO), held in Queensland, Australia.

For the first time, AI systems reached gold-medal-level scores at the prestigious global competition for students under 20. Google’s Gemini and OpenAI’s experimental reasoning model each scored 35 out of a possible 42 points—earning gold-medal equivalents. However, neither could match the five human participants who achieved perfect scores.

Google revealed that its advanced Gemini chatbot solved five out of six problems. “We can confirm that Google DeepMind has reached the much-desired milestone,” said IMO President Gregor Dolinar in a statement shared by the company. “Their solutions were astonishing in many respects—clear, precise, and easy to follow.”

OpenAI also confirmed its model reached the same score. Researcher Alexander Wei stated that each of the AI’s submissions was graded independently by former IMO medalists under standard rules.

Among the 641 contestants from 112 countries, roughly 10 percent earned gold-level scores, but only five humans reached the flawless 42-point mark—something no AI model has yet achieved.

The showing still represents significant progress. In 2024, Google’s system managed only a silver-equivalent score and took days to solve four problems. This year, Gemini completed its solutions within the official 4.5-hour window.

Organisers confirmed that tech firms used the actual 2025 IMO problems in private AI evaluations. However, Dolinar noted that the level of computing resources or possible human involvement in the AI runs could not be verified. “It is very exciting to see progress in the mathematical capabilities of AI models,” he added, while underscoring the continued dominance of top human minds—at least for now.