Alphabet's Google and Openai said their artificial intelligence models are marking a breakthrough in race mathematical capabilities to win gold medals in the global mathematics competition and build powerful systems that rival human intelligence. The results mark the first time that the AI system has crossed the gold medal scoring threshold at the International Mathematics Olympiad in high school students.
Their models solved five of six problems and achieved results using a generic “inference” model that uses natural language to handle mathematical concepts, in contrast to previous approaches used by AI companies.
According to Junyuk Jung, a mathematics professor at Brown University and a researcher visiting researchers at Google's Deepmind AI unit, AI suggests that it has been less than a year since mathematicians were used to solve unsolved research problems on the frontiers of the field.
“The moment we can solve the difficult reasoning problems of natural language, I think the possibility of cooperation between AI and mathematicians will be possible,” Jung told Reuters.
Openai's breakthrough was achieved with a new experimental model focusing on significantly scaling the “test time calculations.” This was done by allowing the model to “think” for a longer period of time, and according to Noam Brown, OpenAI researcher, they simultaneously deployed parallel computing power to simultaneously deploy parallel computing power. Brown refused to say the amount of Openai's cost of computing power, but called it “very expensive.”
For Openai researchers, it is another clear indication that AI models can command a wide range of inference functions that can extend beyond mathematics to other areas.
Jung, who won the IMO Gold Medal in 2003, shares optimism is shared by Google researchers who believe the capabilities of AI models can be applied to research obstacles in other fields such as physics.
Of the 630 students participating in the 66th IMO on the Sunshine Coast in Queensland, Australia, 67 contestants, or about 11%, achieved gold medal scores. Google's Deepmind AI unit achieved a silver medal score last year using a mathematics-specific AI system. This year, Google used a generic model called Gemini Deep Think. That version was previously announced at the annual developer conference in May.
Unlike previous AI attempts that relied on formal languages and long calculations, Google's approach worked perfectly in natural language this year, solving the issue within the official 4.5 hour time limit, the company said in a blog post. According to a post by researcher Alexander Way of social media platform X, Openai, with its own inference model, has similarly built an experimental version for competition.
This year, the first time the competition has been officially coordinated with AI developers has been using prominent mathematics competitions like IMO to test model functionality for many years. The IMO judge recognized the results of those companies, including Google, and asked them to publish the results on July 28th.
“We only shared the results after all AI labs have verified their official results by independent experts, and respected the original request of the IMO committee that students have received the praise they deserved,” Google Deepmind CEO Demis Hassabis said Monday.
Openai, who announced the results on Saturday and first claimed gold medal status, said in an interview that he has permission from IMO board members to do so after Saturday's closing ceremony.
The competition on Monday allowed partners to announce the results, Gregor Driner, president of the IMO board, told Reuters.
Published – July 22, 2025 09:55 AM IST
