OpenAI and Google outdo the mathletes, but not each other

From the AI models Open And Google Dipmind One of the world’s oldest and challenging high school level mathematics competitions, 2025 has earned gold medal scores in the International Math Olympiad (IMO), organizations have announced independently in recent days.

The result refers to how faster the AI systems are moving forward and how the Google and OpenAI AI race seem to be evenly matched. AI companies are intensely competing for public concepts in front of AI race: an indomitable war of “Vobes” that can have a big impact on the protection of top AI talent. Lots of AI researchers come from the background in competitive maths, so criteria like IMO refers to others.

Last year, Got a silver medal in Google IMO Using a “formal” system, which means it is needed to translate problems in a machine -textable format. This year, both OpenAI and Google have entered the “informal” system in the competition, which was able to accept questions and create evidence -based answers in natural language. Both companies claim that their AI models have scored more than most high school students and Google’s AI model than last year, without translating any human-mesin.

In an interview with TechCrunch, researchers have claimed that these gold medal performances are presenting around the AI rational models in non-apparent domains. Although AI rational models are prone to doing good in questions with straight answers such as math or coding functions, these systems fight in tasks such as buying great chairs or helping in complex research with more vague solutions.

However, Google is raising questions about how Open has operated its gold medal IMO performance and announced. After all, if you are about to enter AI models in a math competition for high school, you can argue like adolescents.

Shortly after the Opena announced its glory on Saturday morning Google Dipmind CEO And researchers went on social media Slum Openai’s gold – for the premature declaration of intermediary – Shortly after the IMO announced a high school competition on Friday night – and their model’s examination is not officially evaluated by the IMO.

BTW aside, we did not announce Friday because we respected the main request of the IMO Board that all AI labs were appreciated after their results only after their results were shared by independent experts and students were appropriately appreciated they deserve.

– Demis Hussebas (@Demisabisabis) July 21, 2025

Google Dipmind Senior Researcher and IMO Project -led Thang Lung told TechCrunch that he was waiting to announce his IMO results to honor students who participated in the Google competition.

Lung said that Google has been working with the IMO organizers since last year to prepare for the exam and IMO wanted to stay in the president’s blessing and official grading first Its official results are announced, which it did on Monday morningThe

“IMO organizers have their grading guidelines,” said Lung. “So no assessment based on that guideline could not claim about the gold medal level [performance]The “

Senior OpenAI researcher Nom Brown, working on the IMO model, told TechCrunch that the IMO reached the Open AE a few months ago to participate in a formal mathematics competition, but ChatGPT-producer denied because it was working on the natural language system that it seemed more valuable to follow. Brown said the Open did not know that IMO was conducting an informal exam with Google.

Open says that it has appointed a third-party evaluator-three former IMO medalists who understand the grading system to grad the AI model performance. Asked about his gold medal score, Brown said the company had reached the IMO, which was then asked the agency to wait for the IMO to be announced after the award on Friday night.

IMO did not respond to TechCrunch’s request for comments.

Google is not necessarily wrong here – it has gone through the more official, rigorous process to achieve its gold medal score – but the debate can miss the bigger image: AI models of several top AI labs are rapidly improving. Countries around the world have sent their brilliant students to compete in IMO this year and do a few percent of them as well as OpenAI and Google’s AI models.

Although there is a significant leadership on the opening industry, it certainly seems to be more closely matched if a company wants to acknowledge. Open is expected to release GPT -5 in the coming months, and the company must give the idea that it still leads the AI industry.