AI Gold Medals at IMO Fuel Tech Rivalry

AI models from OpenAI and Google DeepMind achieved gold-medal level performance at the International Math Olympiad, demonstrating a significant leap in AI reasoning. This shared triumph, however, quickly led to a public dispute between the tech giants, underscoring the fierce competition in the AI race.

The rarefied world of high school competitive mathematics, typically a realm of quiet intellectual pursuit and the occasional triumphant roar from a budding genius, has been unexpectedly thrust into the very public, fiercely competitive arena of artificial intelligence.

In a development that signals both staggering technological advancement and the increasingly tribal nature of the AI race, models from OpenAI and Google DeepMind have reportedly clinched gold medals in the 2025 International Math Olympiad (IMO), a contest designed to challenge the brightest young minds on the planet.

This achievement is monumental.

Both companies claim their AI systems independently mastered five out of six incredibly complex problems, a feat that places them above the vast majority of human participants.

Crucially, these were “informal” systems, meaning they could digest questions in natural language and produce proof-based answers without the need for human translation into machine-readable code – a significant leap from Google’s “formal” silver-medal system of last year.

It underscores a breakthrough in AI reasoning, particularly in what researchers term “non-verifiable domains,” where solutions are less straightforward than simple arithmetic or coding tasks.

The ability to navigate ambiguity, to truly reason through a complex, multi-step mathematical proof, marks a profound evolution for these models.

Yet, as impressive as this shared triumph is, the narrative quickly devolved into something less about silicon brilliance and more about silicon squabbles.

Like teenagers jostling for bragging rights after an exam, Google and OpenAI immediately began disputing not the fact of the gold medal, but the manner of its announcement.

The subtext here is palpable: in the high-stakes, talent-hungry world of AI, perception is paramount.

The “vibes” – who is perceived to be ahead, who is more innovative, who attracts the best minds – can have tangible impacts on investment and talent acquisition.

And for many top AI researchers, whose backgrounds often include competitive math, benchmarks like the IMO carry an outsized weight.

The first volley came from Google DeepMind.

Shortly after OpenAI trumpeted its success on a Saturday morning, Google’s CEO and researchers took to social media, accusing their rival of premature celebration.

Thang Luong, a Google DeepMind senior researcher and lead for their IMO project, articulated the company’s position to TechCrunch, stating they deliberately held back their announcement to respect the IMO Board’s request for official verification and, more importantly, to allow the human students their well-deserved moment of acclamation.

Google, having worked with IMO organizers since last year, waited for the IMO president’s blessing and official grading before making its own announcement days later.

Luong emphasized the sanctity of IMO’s grading guidelines, suggesting that any self-evaluation outside that framework couldn’t legitimately claim a “gold-medal level performance.”

It was a subtle but pointed jab at OpenAI’s methodology.

OpenAI, for its part, offered a different perspective.

Noam Brown, a senior OpenAI researcher involved in their IMO model, explained that IMO had indeed reached out to them months prior about participating in a formal competition, which OpenAI declined, prioritizing their natural language systems.

They were, Brown claimed, unaware that an informal test with Google was even underway.

Upon achieving their gold-medal score through evaluation by three former IMO medalists they independently hired – experts who understood the nuances of the grading system – OpenAI says they contacted IMO, who then advised them to wait until after the student awards ceremony.

The IMO itself has remained silent on the matter, declining requests for comment.

This corporate spat, while understandable in the context of the cutthroat AI race, almost overshadows the true significance of the moment.

Who announced first, or whose evaluation process was “more official” in a nascent competitive space, becomes almost a triviality when faced with the sheer scale of the achievement.

Countries scour the globe for their brightest young minds to compete at the IMO, and only a tiny fraction achieve the gold-medal standard.

Now, two artificial intelligences have joined that elite club, demonstrating a level of mathematical reasoning that was, until very recently, considered solely within the domain of human genius.

The deeper truth, often obscured by the public relations skirmishes, is that AI models from multiple leading labs are advancing at an astonishing pace.

OpenAI, once seen as having a significant, even unassailable, lead in the industry, now finds itself in a neck-and-neck race.

Google’s performance at the IMO is just one more piece of evidence that the gap has narrowed considerably, creating an intensely competitive landscape where every perceived advantage, every “vibe,” is fiercely contested.

As OpenAI reportedly gears up for the release of GPT-5 in the coming months, the company undoubtedly hopes to reassert its dominance and reclaim the narrative of being the undisputed leader.

But if the IMO contest is any indication, the battle for AI supremacy is far from over, and the arguments over who crossed the finish line “properly” are likely just the beginning.