Openai and Google surpass mathletes, but not each other

The OPENAI and Google Deepmind IA models obtained scores of gold medals during the International Mathematics (IMO) Olympiad in 2025, one of the oldest and most difficult mathematics competitions in high school, companies announced independently.

The result underlines how much the AI systems progress quickly, and yet, how equal Google and Openai seem to be in the AI race. AI companies are in competition with the public perception of being ahead of the AI breed: an intangible battle of “vibrations” which can have great implications to ensure the best talent of AI. Many IA researchers come from competitive mathematics environments, so benchmarks like the OMI mean more than others.

Last year, Google obtained a silver medal at the OMI using a “formal” system, which means that it forced humans to translate problems in a format readable by machine. This year, Openai and Google entered “informal” systems in competition, which were able to ingest questions and generate answers based on natural language evidence. The two companies claim that their AI models have correctly answered five questions out of the OMI test, marking more than most high school students and the Google IA model from last year, without requiring human machine translation.

In interviews with Techcrunch, the researchers behind Openai and the efforts of the Google OMI claimed that these gold medal performances represent breakthroughs around AI reasoning models in non -veritifiable fields. While AI reasoning models tend to resolve questions with simple answers, such as simple mathematics or coding tasks, these systems fight on tasks with more ambiguous solutions, such as the purchase of a large chair or helping complex research.

However, Google raises questions about how Openai has led and announced its performance as a gold medal. After all, if you are going to enter the AI models in a mathematics competition for high school students, you could just as well argue like adolescents.

Shortly after Openai announced his feat on Saturday morning, the CEO and the researchers of Google Deepmind went to social networks to slam Openai for having announced his stay in gold prematurely – shortly after the OMI announced which high school students had won the competition on Friday evening – and for not being officially evaluated by OMO.

Thang Luong, a main researcher from Google Deepmind and leader for the IMO project, told Techcrunch that Google had waited to announce its OMI results to respect the students participating in the competition.

Luong said Google has been working with the OMI organizers since last year in preparation for the test and wanted to have the blessing and official rating of the OMI president before announcing its official results, which he did on Monday morning.

“OMI organizers have their rating guidelines,” said Luong. “Thus, any assessment that is not based on this directive could not make any complaints on the level of gold medal [performance]. “”

Noam Brown, an Openai Senior researcher who worked on the OMI model, told Techcrunch that Imo had contacted Openai a few months ago on participation in an official mathematics competition, but the chatpt manufacturer refused because he worked on natural language systems that she thought was worth the penalty of being prosecuted. Brown says that Optai did not know that the OMI was doing an informal test with Google.

OPENAI says it has hired third -party assessors – three former OMI medalists who understood the rating system – to note the performance of its AI model. After Openai learned of his gold medal score, Brown said that the company contacted the OMI, who then told the company to wait until the ceremony of the Friday evening of IMO.

The OMI did not respond to the request for comments from Techcrunch.

Google is not necessarily false here – it has passed a more official and rigorous process to reach its gold medal score – but the debate can miss the situation as a whole: the AI models of several main Laboratories of AI improve quickly. Countries around the world have sent their brightest students to participate in the OMI this year, and only a few percent of them have marked as well as the models of IA Openai and Google.

While Optai used to have an important advance on the industry, it certainly seems that the race was more closely equaled than any company wishes. Openai should publish the GPT-5 in the coming months, and the company certainly hopes to give the impression that it always leads the AI industry.