The Gemini Deep Think mode announced at I/O 2025 is not here yet, but Google today is highlighting how it achieved a gold-medal level performance in a math, or maths, competition.
The International Mathematical Olympiad (“IMO”) is the world’s most prestigious competition for young mathematicians, and has been held annually since 1959. Each country taking part is represented by six elite, pre-university mathematicians who compete to solve six exceptionally difficult problems in algebra, combinatorics, geometry, and number theory. Medals are awarded to the top half of contestants, with approximately 8% receiving a prestigious gold medal.
IMO 2025 was held last week, and an “advanced version of Gemini Deep Think solved five out of the six IMO problems perfectly, earning 35 total points, and achieving gold-medal level performance.” Google shared the solutions here (PDF).
To make the most of the reasoning capabilities of Deep Think, we additionally trained this version of Gemini on novel reinforcement learning techniques that can leverage more multi-step reasoning, problem-solving and theorem-proving data. We also provided Gemini with access to a curated corpus of high-quality solutions to mathematics problems, and added some general hints and tips on how to approach IMO problems to its instructions.
Back in May, Google explicitly said Gemini 2.5 Pro was the underlying model. Today’s blog post just says “advanced version” or “advanced Gemini.”
This competition is a good challenge for an AI system’s advanced mathematical problem-solving and reasoning capabilities. In 2024, Google DeepMind scored a silver (“solving four out of the six problems and scoring 28 points”) using AlphaGeometry and AlphaProof with 2-3 days of computation. However, the problems had to first be translated from natural language into domain-specific languages.
This year, Gemini “operated end-to-end in natural language, producing rigorous mathematical proofs directly from the official problem descriptions – all within the 4.5-hour competition time limit.”
Deep Think is an “enhanced reasoning mode” that uses the “latest research techniques,” like parallel thinking.
This setup enables the model to simultaneously explore and combine multiple possible solutions before giving a final answer, rather than pursuing a single, linear chain of thought.
Google says it will make “a version of this Deep Think model available to a set of trusted testers, including mathematicians.” It will come to Google AI Ultra after that, but it’s unclear when it will actually launch to subscribers of the $250 per month tier.
FTC: We use income earning auto affiliate links. More.
Comments