In a historic leap for artificial intelligence, Google DeepMind and OpenAI have achieved gold-medal-level scores at the International Mathematical Olympiad (IMO), a prestigious contest for the world’s top high school mathematicians.
For the first time, AI models not only matched but surpassed the gold medal benchmark, solving five out of six Olympiad problems using general-purpose reasoning models that process mathematical concepts through natural language—breaking away from previous highly specialized AI approaches.
Google DeepMind collaborated directly with IMO officials to have its AI model, Gemini Deep Think, formally graded and certified. OpenAI, while not officially competing, revealed its experimental reasoning model achieved similar gold-medal-worthy results, verified by three former IMO medalists.
“This milestone shows we are less than a year away from AI collaborating with mathematicians on unsolved problems,” said Junehyuk Jung, a Brown University professor and visiting researcher at DeepMind.
OpenAI’s breakthrough relied on scaling “test-time compute,” allowing the model to “think” longer and run parallel reasoning chains simultaneously. While the exact computing cost remains undisclosed, OpenAI researcher Noam Brown called it “very expensive.”
Experts believe this advancement could soon extend beyond mathematics to fields like physics and theoretical research, unlocking AI-human collaboration on frontier problems.
This year’s IMO saw 67 out of 630 students earn gold medals. Google’s win marks a dramatic improvement after last year’s silver-level performance, highlighting how natural language reasoning is transforming AI’s problem-solving potential.
