The AI models from Openai and Google Deepmind achieved gold medal scores at the 2025 International Mathematics Olympiad (IMO), one of the world’s oldest and most challenging high school-level mathematics competitions.
The results highlight how fast the AI system is moving, but still highlight that Google and Openai appear to be in the AI race. AI companies are fiercely competing for the public’s perception that they are ahead of the AI race. This is an intangible battle of “atmosphere” that has a great deal of meaning to secure the talent of top AI. Many AI researchers come from competitive mathematics backgrounds, so benchmarks like IMO make more sense than others.
Last year, Google won a silver medal at IMO using its “official” system. In other words, humans had to convert the problem into a machine-readable format. This year, both Openai and Google have taken the “unofficial” system into the competition. This allowed us to ingest questions and generate proof-based answers in natural language. Both companies claim that the AI model correctly answered five of the six questions about the IMO testing, scoring higher than most high school students and Google’s AI models last year.
In an interview with TechCrunch, researchers behind Openai and Google’s IMO efforts argued that the performance of these gold medals represents a breakthrough centered around AI inference models in the unverified domain. AI inference models tend to do well with questions with simple answers, such as simple mathematics and coding tasks, but these systems struggle with tasks with more ambiguous solutions, such as buying a good chair or supporting complex research.
However, Google has raised questions about how Openai performed and announced the gold medal IMO performance. After all, if you’re going to participate in AI models in a mathematics contest for high school students, you can also discuss it like a teenager.
Shortly after Openai announced the feat on Saturday morning, Google Deepmind CEOs and researchers took Slam Openai to social media to slam the gold medal early after IMO announced which high school students won the competition on Friday night.
Thang Luong, a senior researcher at Google Deepmind and lead of the IMO project, told TechCrunch that he is waiting for Google to release IMO results to honor students who are competing.
TechCrunch Events
San Francisco
|
October 27th-29th, 2025
Luong said Google has been working with IMO organizers to prepare for the test since last year and wanted the IMO president’s blessing and official grading before releasing official results on Monday morning.
“IMO organizers have grading guidelines,” says Luong. “Therefore, ratings that were not based on that guidelines could not be argued about gold medal levels. [performance]. ”
Norm Brown, a senior open researcher who worked on the IMO model, told TechCrunch a few months ago that IMO had reached out to Openai about participating in a formal mathematics competition, but declined because ChatGpt-Maker was working on a natural language system and thought it was worth pursuing. Brown says Openai didn’t know that IMO was doing unofficial testing on Google.
Openai says it hired three former IMO medalists who understood the grading system to assess the performance of the AI model. After Openai learned of the gold medal score, Brown told the company that the company contacted the IMO and waited for the announcement until after the IMO Friday night awards ceremony.
IMO did not respond to TechCrunch’s request for comment.
Google isn’t necessarily wrong. It goes through a more formal and rigorous process to achieve a gold medal score, but the discussion may miss a bigger picture. Countries around the world were sent to compete with the brightest students of the year at IMO, scoring only a few percent of them, as were the AI models from Openai and Google.
Openai once had a more important lead than the industry, but that certainly feels like the race is more closely aligned than any company wants to acknowledge. Openai plans to release the GPT-5 in the coming months and wants to reveal the impression that the company is certainly leading the AI industry.
Source link
