OpenAI and Google outdo the mathletes, however not one another

AI fashions from OpenAI and Google DeepMind achieved gold-medal scores within the 2025 Worldwide Math Olympiad (IMO), one of many world’s oldest and most difficult excessive school-level math competitions, the businesses independently introduced in latest days.

The outcomes underscore simply how briskly AI techniques are advancing, and but, how evenly matched Google and OpenAI appear to be within the AI race. AI firms are competing fiercely for the general public notion of being forward within the AI race: an intangible battle of “vibes” that may have massive implications for securing prime AI expertise. A whole lot of AI researchers come from backgrounds in aggressive math, so benchmarks like IMO imply greater than others.

Final yr, Google scored a silver medal at IMO utilizing a “formal” system, which means it required people to translate issues right into a machine‑readable format. This yr, each OpenAI and Google entered “casual” techniques into the competitors, which have been capable of ingest questions and generate proof‑primarily based solutions in pure language. Each firms declare their AI fashions accurately answered 5 out of six questions on IMO’s check, scoring increased than most highschool college students and Google’s AI mannequin from final yr, with out requiring any human-machine translation.

In interviews with TechCrunch, researchers behind OpenAI and Google’s IMO efforts claimed that these gold-medal performances symbolize breakthroughs round AI reasoning fashions in non-verifiable domains. Whereas AI reasoning fashions are inclined to do nicely on questions with simple solutions, equivalent to simple arithmetic or coding duties, these techniques wrestle on duties with extra ambiguous options, equivalent to shopping for an ideal chair or serving to with advanced analysis.

Nonetheless, Google is elevating questions round how OpenAI performed and introduced its gold-medal IMO efficiency. In spite of everything, for those who’re going to enter AI fashions right into a math contest for prime schoolers, you may as nicely argue like youngsters.

Shortly after OpenAI introduced its feat on Saturday morning, Google DeepMind’s CEO and researchers took to social media to slam OpenAI for saying its gold medal prematurely — shortly after IMO introduced which excessive schoolers had received the competitors on Friday evening — and for not having their mannequin’s check formally evaluated by IMO.

Thang Luong, a Google DeepMind senior researcher and lead for the IMO challenge, informed TechCrunch that Google waited to announce its IMO outcomes to respect the scholars collaborating within the competitors.

Luong stated that Google has been working with IMO’s organizers since final yr in preparation for the check and wished to have the IMO president’s blessing and official grading earlier than saying its official outcomes, which it did on Monday morning.

“The IMO organizers have their grading guideline,” Luong stated. “So any analysis that’s not primarily based on that guideline couldn’t make any declare about gold-medal stage [performance].”

Noam Brown, a senior OpenAI researcher who labored on the IMO mannequin, informed TechCrunch that IMO reached out to OpenAI a number of months in the past about collaborating in a proper math competitors, however the ChatGPT-maker declined as a result of it was engaged on pure language techniques that it thought have been extra price pursuing. Brown says OpenAI didn’t know IMO was conducting a casual check with Google.

OpenAI says it employed third-party evaluators — three former IMO medalists who understood the grading system — to grade its AI mannequin’s efficiency. After OpenAI discovered of its gold-medal rating, Brown stated the corporate reached out to IMO, which then informed the corporate to attend to announce till after IMO’s Friday evening award ceremony.

IMO didn’t reply to TechCrunch’s request for remark.

Google isn’t essentially fallacious right here — it did undergo a extra official, rigorous course of to attain its gold-medal rating — however the debate could miss the larger image: AI fashions from a number of main AI labs are enhancing rapidly. Nations from around the globe despatched their brightest college students to compete at IMO this yr, and only a few % of them scored in addition to OpenAI and Google’s AI fashions did.

Whereas OpenAI used to have a major lead over the business, it definitely feels as if the race is extra intently matched than any firm want to admit. OpenAI is anticipated to launch GPT-5 within the coming months, and the corporate definitely hopes to offer off the impression that it nonetheless leads the AI business.