The International Mathematical Olympiad (IMO) is one of the most prestigious competitions for mathematicians around the world. In this competition, each country is represented by six pre-university students who compete to solve six exceptionally difficult problems. Each year, only about 8% of them receive the prestigious gold medal.
At this year"s IMO, both OpenAI and Google used their latest LLMs to compete. Over the weekend, OpenAI claimed that its experimental reasoning LLM achieved a gold medal-level performance. OpenAI evaluated its models on the 2025 IMO problems under the same rules as students. The models had to solve the problems in two 4.5-hour exam sessions, with no access to tools or the internet, and write natural language proofs.
We achieved gold medal-level performance 🥇on the 2025 International Mathematical Olympiad with a general-purpose reasoning LLM!
— OpenAI (@OpenAI) July 19, 2025
Our model solved world-class math problems—at the level of top human contestants. A major milestone for AI and mathematics. https://t.co/u2RlFFavyT
Today, Google claimed that an advanced version of Gemini Deep Think solved five out of the six IMO problems, achieving gold-medal-level performance. This advanced version of Gemini Deep Think is powered by an improved reasoning mode using Google DeepMind"s latest research work, including parallel thinking. This enabled Gemini to explore multiple solutions simultaneously instead of exploring a single solution one by one.
The Google DeepMind team wrote the following regarding this advanced version of Gemini Deep Think:
"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."
Google will be releasing this new version of the Deep Think model to a set of trusted testers in the near future. Later, it will also be made available to Google AI Ultra subscribers.