OpenAI and Google Advance AI in Mathematics, Yet Rivalry Persists

AI models from OpenAI and Google DeepMind have achieved gold-medal performance at the 2025 International Math Olympiad (IMO), one of the world's most prestigious and difficult high school mathematics competitions, as both companies recently announced.

These results highlight the rapid progress of AI systems while revealing how closely matched Google and OpenAI remain in the AI race. The competition between AI companies extends beyond technology to public perception—a battle of "vibes" that significantly influences their ability to attract top AI talent. Since many AI researchers have competitive mathematics backgrounds, benchmarks like IMO carry particular weight in this field.

Last year, Google earned a silver medal using a "formal" system that required human translation of problems into machine-readable format. This year, both companies submitted "informal" systems capable of processing questions directly and generating proofs in natural language. Both claim their AI models correctly solved five of the six competition problems, outperforming most human participants and Google's previous AI system, all without human-assisted translation.

In discussions with TechCrunch, researchers from both companies described these gold-medal achievements as breakthroughs in AI reasoning for non-verifiable domains. While AI systems typically excel at tasks with clear answers like basic math or coding, they've historically struggled with ambiguous challenges such as selecting optimal furniture or assisting with complex research.

However, Google has raised questions about OpenAI's approach to announcing its IMO results. After all, when entering AI models in a high school mathematics competition, some adolescent-style squabbling seems almost expected.

Shortly after OpenAI's Saturday morning announcement, Google DeepMind's CEO and researchers used social media to criticize OpenAI for declaring victory prematurely—following Friday night's announcement of student winners—and for lacking official IMO evaluation of their model's performance.

As a relevant note, we delayed our Friday announcement to respect the IMO Board's request that AI laboratories share results only after official verification by independent experts and after the students received proper recognition for their achievements.

— Demis Hassabis (@demishassabis) July 21, 2025

Thang Luong, Google DeepMind's senior researcher leading the IMO project, explained to TechCrunch that Google delayed its announcement to honor the student competitors.

Tech and VC heavyweights join the Disrupt 2025 agenda

Netflix, ElevenLabs, Wayve, Sequoia Capital — just a few of the heavy hitters joining the Disrupt 2025 agenda. They’re here to deliver the insights that fuel startup growth and sharpen your edge. Don’t miss the 20th anniversary of TechCrunch Disrupt, and a chance to learn from the top voices in tech — grab your ticket now and save up to $675 before prices rise.

Tech and VC heavyweights join the Disrupt 2025 agenda

Netflix, ElevenLabs, Wayve, Sequoia Capital — just a few of the heavy hitters joining the Disrupt 2025 agenda. They’re here to deliver the insights that fuel startup growth and sharpen your edge. Don’t miss the 20th anniversary of TechCrunch Disrupt, and a chance to learn from the top voices in tech — grab your ticket now and save up to $675 before prices rise.

Luong noted that Google collaborated with IMO organizers throughout the past year in test preparation, seeking official approval from the IMO president and formal grading before announcing their verified results on Monday morning.

"The IMO organizers maintain specific grading guidelines," Luong emphasized. "Any evaluation not following those standards cannot legitimately claim gold-medal level performance."

Noam Brown, a senior OpenAI researcher involved in the IMO project, told TechCrunch that IMO contacted OpenAI months earlier about participating in a formal mathematics competition, but the ChatGPT creator declined, focusing instead on natural language systems it considered more promising. Brown stated OpenAI wasn't aware IMO was conducting informal testing with Google.

OpenAI reports hiring third-party evaluators—three former IMO medalists familiar with the grading system—to assess their AI model's performance. After learning of their gold-medal score, Brown said the company contacted IMO, which then requested they delay announcement until after Friday night's award ceremony.

IMO has not responded to TechCrunch's request for comment.

While Google's concerns about proper procedure have merit—having undergone a more official and rigorous evaluation process—the controversy may obscure a more significant development: AI models from leading laboratories are advancing rapidly. Among the world's brightest mathematics students gathered at this year's IMO, only a small percentage matched the performance achieved by OpenAI and Google's AI systems.

Though OpenAI previously maintained a clear industry lead, the competition now appears much tighter than any company willingly acknowledges. With GPT-5 expected in coming months, OpenAI undoubtedly hopes to reinforce its position as the AI industry frontrunner.