Google DeepMind has announced that its advanced Gemini model has officially achieved gold-medal performance at IMO 2025, following OpenAI’s similar claim.

Tested with the same time constraints and problems as humans, Gemini solved 5 out of 6 problems, scoring 35/42—the gold standard. Unlike OpenAI, DeepMind collaborated directly with IMO for validation, and its answers were graded using official criteria by IMO coordinators.

This year marks a leap forward as the model worked fully end-to-end in natural language. While OpenAI’s claims sparked skepticism due to lack of official evaluation, DeepMind’s transparency adds credibility. Both show frontier AI is nearing top-tier math capabilities, though none have yet solved all 6 problems.