Google DeepMind's advanced Gemini Deep Think model has achieved gold medal-level performance at the 2025 International Mathematical Olympiad (IMO), scoring 35 out of 42 points by solving five of six problems perfectly. The achievement was officially certified by IMO coordinators using the same grading criteria applied to student participants.
The IMO President Prof. Dr. Gregor Dolinar confirmed that "Google DeepMind has reached the much-desired milestone, earning 35 out of a possible 42 points — a gold medal score. Their solutions were astonishing in many respects. IMO graders found them to be clear, precise and most of them easy to follow."
The International Mathematical Olympiad, held annually since 1959, represents the world's most prestigious competition for young mathematicians. Students solve six exceptionally difficult problems across algebra, combinatorics, geometry, and number theory, with approximately 8% of contestants earning gold medals.
This year's result marks a significant technical advancement over Google's 2024 performance, when its AlphaProof and AlphaGeometry systems achieved silver medal status with 28 points. The 2025 model operated end-to-end in natural language, producing rigorous mathematical proofs directly from official problem descriptions within the 4.5-hour competition time limit, eliminating the need for manual translation into formal mathematical languages.
The system used an enhanced version of Gemini Deep Think that incorporates parallel thinking capabilities, allowing it to simultaneously explore multiple solution paths before providing final answers. Google trained this version using novel reinforcement learning techniques and provided access to curated high-quality mathematical solutions.
The breakthrough has drawn attention from the AI research community, particularly after OpenAI announced similar results without participating in the official IMO evaluation process. Google's measured approach to releasing results after official verification has been praised by researchers, with DeepMind CEO Demis Hassabis noting they "respected the IMO Board's original request that all AI labs share their results only after the official results had been verified by independent experts".
For developers, this advancement demonstrates AI systems' growing capability to handle complex reasoning tasks using natural language inputs. The progression from requiring specialized formal languages to operating entirely in natural language suggests that AI systems are becoming more intuitive and accessible.
Google plans to make a version of the Deep Think model available to trusted testers, including mathematicians, before rolling it out to Google AI Ultra subscribers at $249.99 per month. The company indicated this could enable mathematicians to tackle increasingly challenging problems with AI assistance.
The single problem the system failed to solve involved finding the minimum number of rectangles needed to cover a given space, which researchers noted was objectively the hardest question, with only five human students solving it correctly.