By Atoyebi Nike
Generative AI models from Google and OpenAI reached gold-level scores for the first time at the 2025 International Mathematical Olympiad (IMO), but human contestants still outperformed them.
Held this month in Queensland, Australia, the IMO saw five young mathematicians achieve perfect scores, while AI models fell just short. Google’s Gemini chatbot solved five of six problems, earning 35 out of 42 points a gold medal score, according to IMO president Gregor Dolinar.
“Their solutions were astonishing… clear, precise, and mostly easy to follow,” Dolinar noted.
OpenAI reported a similar result for its experimental reasoning model, also scoring 35 points. “This achieved a longstanding grand challenge in AI,” said OpenAI researcher Alexander Wei.
Both AI models were evaluated under the same rules as the 641 contestants from 112 countries. Each problem was graded by three former IMO medalists.
While the IMO acknowledged the impressive progress of AI in mathematical reasoning, it noted that human participants still set the bar with faster, perfect solutions. Google’s model completed its test in the standard 4.5-hour limit this year, compared to multiple days of computation in 2024.
However, competition officials cautioned they could not verify how much computing power was used by the companies or whether there was human input.
