OpenAI and Google DeepMind Claim AI Gold at Math Olympiad - Then Clash Over Validation

OpenAI and Google DeepMind Claim AI Gold Medals at Math Olympiad—And Immediately Start Arguing
By [Your Name] | July 22, 2025

In a milestone for AI reasoning, systems from OpenAI and Google DeepMind have independently scored gold-medal results at the 2025 International Math Olympiad (IMO), the world’s most grueling high-school math competition. But within hours of OpenAI’s announcement, the collaboration turned to confrontation—with Google accusing its rival of jumping the gun and violating the competition’s spirit.

The Breakthrough

Both companies revealed their models solved 5 of 6 Olympiad-level problems—a feat achieved by fewer than 4% of human contestants this year. Crucially, their 2025 entries required no human translation: the AIs ingested natural-language problems (e.g., "Prove that for all integers 𝑛 ≥ 3, there exists a real number 𝑐 such that...") and output full proofs autonomously.

This marks a leap from 2024, when Google’s silver-medal system needed formal logic hand-coding by humans. "Informal reasoning is the frontier," said OpenAI’s Noam Brown. "These problems demand creative leaps, not just pattern matching."

The Beef

The dispute erupted minutes after OpenAI’s Saturday announcement:

• Google DeepMind CEO Demis Hassabis tweeted: "Respect the students. Official grading exists for a reason."

• Lead researcher Thang Luong told TechCrunch: "We coordinated with IMO for months. Any claim without their rubric isn’t gold-medal valid."

Google had waited until Monday to announce identical results, citing deference to the Friday-night award ceremony for human winners. OpenAI countered that it used three ex-IMO medalists as graders and claimed IMO organizers instructed them to delay announcing until after the ceremony—which they did.

"IMO approached us about a formal test, but we declined to focus on natural-language systems," Brown said. "We didn’t know Google was doing an informal track with IMO."