Alphabet’s Google and OpenAI have announced that their AI models earned gold medals at a prestigious international mathematics competition, marking a significant breakthrough in AI’s mathematical reasoning abilities.
For the first time, AI systems surpassed the gold-medal scoring threshold at the International Mathematical Olympiad (IMO), a competition for high school students. Both companies’ models solved five out of six challenging problems using general-purpose reasoning systems that understand math concepts through natural language, differing from previous AI approaches.
This milestone suggests AI could be less than a year away from assisting mathematicians in solving unsolved problems at the cutting edge of the discipline, according to Junehyuk Jung, a Brown University math professor and visiting researcher at Google’s DeepMind. Jung, himself a former IMO gold medalist, also mentioned that such AI collaboration could extend to research challenges in other fields like physics.
At the 66th IMO held in Queensland, Australia, 67 of the 630 student participants (about 11%) earned gold medals. Last year, Google’s DeepMind achieved a silver medal score with specialized AI for math. This year, Google used a general-purpose model called Gemini Deep Think, which solved problems entirely through natural language within the 4.5-hour time limit. OpenAI developed a similar reasoning model for the competition but has not yet announced plans to release it publicly.
This was the first year the IMO officially coordinated with some AI developers, who have used the competition for years to benchmark their systems. IMO judges verified and certified the AI results, asking the companies to wait until July 28 to publish them. Google’s DeepMind CEO Demis Hassabis said they respected the IMO Board’s request to prioritize recognition of the student participants.
However, OpenAI, which did not collaborate directly with the IMO, released its results earlier, becoming the first AI firm to claim gold-medal status. Following this, the IMO allowed cooperating companies to publish their verified results publicly.
