Gemini Deep Think reaches gold-medal standard at IMO 2025
An advanced Gemini system produced natural-language solutions to five of six olympiad problems within the competition time limit, with scoring confirmed by IMO coordinators.
What was reported
Google DeepMind reported that an advanced version of Gemini Deep Think solved five of the six 2025 International Mathematical Olympiad problems, scoring 35 out of 42 points under official grading criteria.
Unlike the specialist formal pipeline used by AlphaProof in 2024, the 2025 system worked directly from the natural-language problem statements and returned natural-language proofs within the standard contest time limit.
Why it matters
Olympiad problems are not open research problems, but the result provides a concrete public reference point for the progression from formal specialist systems toward general-purpose mathematical reasoning. The linked report includes the organizers' attribution and access to the submitted solutions.
OpenTCS reports the public result without introducing a separate site-defined verification grade.