Gemini Deep Think achieves gold-medal standard at IMO 2025

Gemini Deep Think reaches gold-medal performance at IMO 2025

An advanced version of Gemini with Deep Think has officially achieved a gold-medal standard at the 2025 International Mathematical Olympiad (IMO), solving five out of six problems perfectly for a total of 35 points. This result was officially graded and certified by IMO coordinators using the same criteria applied to student solutions.

Technical evolution from AlphaProof and AlphaGeometry

This achievement represents a significant leap in AI mathematical reasoning compared to the 2024 results. In 2024, Google DeepMind's AlphaProof and AlphaGeometry 2 systems achieved a silver-medal standard by solving four problems and scoring 28 points, but required expert translation of problems into domain-specific languages like Lean and took two to three days of computation.

In contrast, the 2025 Gemini Deep Think 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.

The Deep Think reasoning architecture

Gemini Deep Think is an enhanced reasoning mode designed for complex problems. It utilizes several key research techniques to improve problem-solving capabilities:

  • Parallel Thinking: Instead of following a single linear chain of thought, the model can simultaneously explore and combine multiple possible solution paths before arriving at a final answer.
  • Reinforcement Learning: The model was trained using novel reinforcement learning techniques specifically designed to leverage multi-step reasoning, theorem-proving, and problem-solving data.
  • Curated Knowledge: The system was provided with a curated corpus of high-quality mathematical solutions and specific instructions containing general hints and tips for approaching IMO-level problems.

Implications for the future of mathematics and AGI

Google DeepMind views this result as an early step toward building AI that can solve more complex and advanced mathematics. While the 2025 success was based on natural language fluency, the lab continues to develop formal systems like AlphaGeometry and AlphaProof.

DeepMind posits that the future of mathematical discovery lies in agents that combine natural language fluency with verified reasoning in formal languages. Such tools are intended to assist mathematicians, scientists, and engineers in advancing human knowledge on the path to Artificial General Intelligence (AGI).

Availability

A version of the Deep Think model will be released to a set of trusted testers, including mathematicians, before being rolled out to Google AI Ultra subscribers.

Sources