Gemini Deep Think achieves gold-medal standard at International Mathematical Olympiad

Google DeepMind announced that an advanced version of Gemini Deep Think has officially achieved gold-medal standard at the International Mathematical Olympiad (IMO) 2025, solving five out of six problems and earning 35 total points. The results were officially graded and certified by IMO coordinators using the same criteria applied to student competitors.

From Silver to Gold#

This marks a substantial advance over last year's breakthrough. In 2024, the combined AlphaProof and AlphaGeometry 2 systems achieved silver-medal performance by solving four problems and scoring 28 points—but they required expert translation of problems from natural language into domain-specific languages like Lean, plus two to three days of computation.

This year's result demonstrates a fundamental shift: Gemini Deep Think operated end-to-end in natural language, producing rigorous mathematical proofs directly from the official problem descriptions and completing the task within the 4.5-hour competition time limit.

How Deep Think Works#

The breakthrough relied on an enhanced reasoning mode that incorporates parallel thinking—allowing the model to simultaneously explore and combine multiple possible solutions before arriving at a final answer, rather than following a single linear chain of thought.

DeepMind also employed novel reinforcement learning techniques trained on multi-step reasoning and problem-solving data, and provided the model with access to curated high-quality mathematics solutions and general IMO problem-solving guidance.

The Road Ahead#

DeepMind plans to release a version of Deep Think to trusted testers—including mathematicians—before rolling it out to Google AI Ultra subscribers. The team emphasizes that this is only the beginning of AI's potential in mathematics.

"By teaching our systems to reason more flexibly and intuitively, we are getting closer to building AI that can solve more complex and advanced mathematics," DeepMind stated. The team continues advancing both natural-language and formal-language systems, envisioning future agents that combine fluency with rigorous verified reasoning to support mathematicians, scientists, and researchers on the path toward AGI.