Artificial intelligence has taken a significant leap forward, rivaling the brightest young mathematicians at the prestigious International Mathematical Olympiad (IMO).
Google DeepMind’s advanced Gemini model, now featuring “Deep Think” capabilities, stunned the mathematical community by matching gold-medal standards in 2025. This milestone not only showcases AI’s evolving prowess but also hints at a future where machines and humans collaborate to solve complex problems.
Gold-Medal Performance at the IMO
Gemini Deep Think distinguished itself by solving five out of six highly challenging IMO problems, earning 35 out of 42 points a score that places it among the top 8% of human contestants.
The clarity, rigor, and accessible explanations of its solutions were praised by the IMO President, who confirmed they met the competition’s stringent standards. This achievement marks a major advance over prior AI models, which struggled to reach the same level of mathematical sophistication.
From Assisted Proofs to Independent Reasoning
Previous AI systems, such as AlphaProof and AlphaGeometry 2, achieved silver-medal results and required extensive expert input to translate between natural and formal mathematical languages.
These models were slow, taking days to solve problems. In stark contrast, Gemini Deep Think operates entirely in natural language, delivering fully developed, human-readable proofs within the official 4.5-hour contest window, without the need for translation or manual intervention.
Inside Deep Think: How Gemini Succeeds
The breakthrough comes from Gemini’s “Deep Think” mode, which enables parallel thought processes. By exploring several solution paths at once and synthesizing insights, the model can arrive at robust, creative answers. Its impressive performance is powered by:
- Reinforcement learning guided by multi-step reasoning and expert theorem-proving data
- A carefully curated set of mathematical solutions for specialized training
- Custom hinting strategies tailored to the unique challenges of IMO problems
This sophisticated approach allows Gemini to handle the entire range of IMO questions, demonstrating adaptability and depth rarely seen in machine reasoning.
Collaboration and Certification
Developing Gemini Deep Think was a team effort, uniting researchers, engineers, and mathematicians. The system’s results underwent official grading and certification by IMO coordinators, adding an extra layer of credibility. Before broader release, trusted mathematicians will further test the system, with plans for availability to Google AI Ultra subscribers in the near future.
Transforming the Future of Math and AI
While Gemini’s gold-medal showing is a landmark, DeepMind emphasizes that it is only the beginning. The ongoing goal is to merge natural language fluency with formal, verifiable reasoning, creating AI agents that not only solve problems but also communicate solutions effectively. This blend promises to assist professionals in mathematics, science, and research, accelerating discoveries and moving closer to artificial general intelligence (AGI).
A Collective Achievement
The success of Gemini Deep Think is attributed to a diverse, interdisciplinary team whose expertise spanned technical development and mathematical insight. The IMO’s thorough review confirmed the correctness of Gemini’s solutions, though the focus remained on answer validity rather than the model’s inner workings.
Key Takeaway
Gemini Deep Think’s triumph at the IMO marks a transformative moment in AI-driven mathematics. As these systems grow increasingly capable, they are poised to amplify human ingenuity and spark innovation across disciplines. The gold-medal achievement not only proves AI’s advancing abilities but also paves the way for deeper collaboration between technology and the mathematical sciences.
Source: Google DeepMind Blog
Gemini Deep Think: AI Achieves Gold-Medal Status at the International Mathematical Olympiad