AI in Mathematics: Gemini with DeepThinking Sets New Standard at International Mathematical Olympiad

A significant milestone in artificial intelligence (AI) has been reached as an advanced version of Google DeepMind's Gemini, enhanced with "DeepThink" capabilities, officially achieved a gold-medal standard at the International Mathematical Olympiad (IMO). [1] This achievement, confirmed by IMO coordinators, marks the first time an AI system has been formally recognized at this elite level of mathematical competition, demonstrating remarkable progress in AI's reasoning and problem-solving abilities. [1, 2]

Breakthrough at IMO 2025

In a landmark event at the 2025 International Mathematical Olympiad, Google DeepMind's Gemini DeepThink system flawlessly solved five out of six highly challenging problems, accumulating 35 points out of a possible 42. [1] This score is equivalent to a prestigious gold medal, a feat achieved by only 67 out of 630 human participants in the same competition. [1, 2]

What is the International Mathematical Olympiad (IMO)?

The International Mathematical Olympiad (IMO) is the world's most prestigious annual competition for pre-university students, challenging them with exceptionally difficult problems in algebra, combinatorics, geometry, and number theory.

The IMO is recognized globally as the ultimate test of advanced mathematical problem-solving for young mathematicians. Unlike previous AI attempts that required problems to be translated into formal languages, Gemini DeepThink operated end-to-end in natural language, producing rigorous mathematical proofs directly from the official problem descriptions within the competition's 4.5-hour time limit. [1] This performance represents a substantial leap from Google DeepMind’s previous silver-medal standard achieved at IMO 2024 with its AlphaGeometry and AlphaProof systems, which needed expert translation and took significantly longer computation times. [1, 2]

How DeepThink Elevates Gemini's Reasoning

The advanced capabilities demonstrated by Gemini DeepThink stem from its enhanced reasoning mode, designed specifically for complex problems. [1] This mode integrates cutting-edge research techniques, including "parallel thinking." [1]

What is DeepThink? 

DeepThink is an enhanced reasoning mode for Google DeepMind's Gemini model that incorporates advanced research techniques, such as parallel thinking, to simultaneously explore and combine multiple potential solutions for complex problems. [1]

Parallel thinking enables the model to explore and combine multiple possible solutions concurrently, rather than pursuing a single, linear chain of thought. To maximize these reasoning capabilities, this version of Gemini was further trained using novel reinforcement learning techniques. These techniques leverage extensive multi-step reasoning, problem-solving, and theorem-proving data. Additionally, Gemini was equipped with a curated corpus of high-quality mathematics solutions and general strategic advice for tackling IMO problems, further enhancing its ability to generate sophisticated solutions. [1]

Implications for AI and Mathematics

The gold-medal achievement by Gemini DeepThink at the IMO underscores AI's rapidly advancing ability to perform complex, abstract reasoning, signaling a new era for AI's role in scientific discovery and problem-solving. [1, 2]

This success is seen by experts as a significant step towards the use of AI in advanced mathematical research, demonstrating that general-purpose AI models are approaching human-level proficiency in competition mathematics. While competition mathematics differs from open-ended research mathematics, the ability of Gemini to operate autonomously and communicate in natural language is a critical development. [1] Google DeepMind plans to make a version of this DeepThink model available to a select group of trusted testers, including mathematicians, before its broader rollout to Google AI Ultra subscribers. [2] This phased release will allow for further evaluation and integration into research and educational settings. The continuous progress by Google DeepMind highlights the rapid advancements in AI reasoning capabilities and suggests a future where AI agents, combining natural language fluency with rigorous reasoning, will become invaluable tools for mathematicians, scientists, engineers, and researchers, helping to advance human knowledge. [1, 2]

Why This Matters

Gemini DeepThink’s IMO gold medal validates AI’s growing capacity for complex reasoning. This development points toward a future where AI can accelerate scientific discovery and assist in solving some of humanity's most challenging problems, transforming fields beyond just mathematics. For organizations, particularly those in research and development, this milestone signifies the potential for integrating advanced AI into complex problem-solving workflows. It highlights the strategic advantage of investing in models with sophisticated reasoning capabilities, enabling new approaches to innovation and potentially reducing development cycles for highly intricate tasks.