The 2025 International Math Olympiad (IMO) just became the latest battleground in the high-stakes rivalry between OpenAI and Google DeepMind — and both labs walked away claiming gold-medal glory.
In a historic moment for artificial intelligence, both tech giants announced that their AI systems matched the performance of the world’s brightest teenage mathematicians, correctly solving five out of six IMO problems — an achievement that only a tiny percentage of human competitors manage each year. What makes the feat more impressive? These weren’t “formal” math systems requiring structured input and coded prompts. These were “informal” natural language models that understood raw IMO questions and wrote full, proof-based answers in everyday English.
It’s a major milestone in AI’s progress toward mastering open-ended reasoning — and it might just shake up the way we think about intelligence itself.
From Code to Calculus: The Leap Forward
Last year, Google’s system used a formal interface that still required human interpretation to present math problems in a machine-readable way. This year, both labs went full send with natural language models — systems that can take in complex, nuanced prompts and reason through multi-step solutions like a human student might.
But beyond the impressive math scorecards, the subtext is even louder: the race to lead the next generation of AI is tighter than ever, and both labs are eager to plant their flag at the summit of human-like reasoning.
The (Teenage) Drama: Disputes Over Timing and Transparency
OpenAI was first to announce its achievement — on Saturday morning, just hours after IMO revealed the results of the student competition on Friday night. Google responded swiftly, criticizing the timing of OpenAI’s announcement and questioning the credibility of their evaluation process.
“The IMO organizers have their grading guideline,” said Thang Luong, lead researcher on Google DeepMind’s IMO project. “Any evaluation that’s not based on that guideline can’t claim gold-medal performance.”
According to Google, it worked directly with the IMO committee to ensure the evaluation of its model followed official grading procedures. OpenAI, on the other hand, says it recruited three former IMO medalists as independent reviewers to score its model’s answers — and claims it respected IMO’s request to delay its announcement until after the awards ceremony.
Google formally released its results on Monday morning, with full endorsement from the IMO president.
What’s Really at Stake: Talent, Perception, and the Future of Reasoning
While the public spat might seem petty — even adolescent — it reflects a deeper truth in today’s AI landscape. The perception of being “in the lead” matters. Top AI researchers, many of whom come from Olympiad backgrounds, gravitate toward labs seen as cutting-edge. Wins like these aren’t just bragging rights; they’re recruitment tools.
In the broader picture, the ability for AI to reason in ambiguous, multi-step domains — not just recall or compute — is becoming the next frontier. Solving IMO problems isn’t just about math; it’s about understanding, logic, and abstract thought — things long believed to be distinctly human.
A Gold Rush, Not Just a Gold Medal
The fact that two separate AI labs could solve most of the IMO test in natural language — with no human translation layer — is a signal of how far, and how fast, this technology is advancing. It also suggests that the era of a single leader in AI may be over. OpenAI once enjoyed a clear lead in the public imagination; now, Google is gaining ground with rigor and scale.
As OpenAI preps for the likely release of GPT-5, and as Google pushes deeper with its Gemini and AlphaCode initiatives, it’s clear we’re entering a new phase of AI competition — one defined not just by power, but by subtlety, rigor, and trust.
One thing is certain: if AI can ace the toughest high school math test on Earth, then the future of education, talent, and reasoning may never look the same again.
👇 TL;DR (For Your Social Feed)
-
OpenAI and Google DeepMind both claim gold-medal-level scores on the 2025 IMO.
-
Both models were “informal” systems — reading and solving problems in natural language.
-
Google accuses OpenAI of jumping the gun with a premature announcement.
-
OpenAI used independent reviewers; Google worked with IMO organizers directly.
-
The bigger picture: AI is rapidly mastering open-ended reasoning — and the race between labs is heating up.
