In an unprecedented milestone for artificial intelligence, Google’s Gemini Deep Think and a new experimental OpenAI model both achieved gold medal scores at the 66th International Mathematical Olympiad (IMO), held July 2025 in Queensland, Australia. This marks the first time AI systems have reached this level in the prestigious high-school competition (Reuters).
Both systems solved 5 out of 6 IMO problems, each scoring the 35-point threshold required for gold. Unlike previous math AI models that relied on formal or symbolic techniques, these systems tackled problems using natural-language reasoning — mirroring how humans think and communicate (Reuters).
🧠 Google’s Approach: Gemini Deep Think
Operated within the 4.5-hour competition window like human participants.
Solved the problems entirely using natural-language reasoning.
Officially submitted its solutions to the IMO, which were certified by judges (Reuters).
🚀 OpenAI’s Strategy
Employed a novel experimental model optimized for intense “test-time compute,” enabling parallel and deeper reasoning efforts (Reuters).
Evaluated internally using official IMO problems and validated by three independent IMO gold medalists — though not formally entered in the competition (Reuters).
OpenAI plans to release similar high-caliber models in several months, per researcher Alexander Wei (Reuters).
Why This Matters
Only 67 out of 630 human participants earned gold — about 11%(Reuters).
Brown University professor and former IMO gold medalist Junehyuk Jung emphasizes this as a critical shift: AI can now solve “hard reasoning problems in natural language,” paving the way for future collaboration with mathematicians (Reuters).
There’s optimism that such AI reasoning systems could soon tackle unsolved problems in math, physics, and other scientific domains (Reuters, Reuters).
Key Takeaways
Theme
Insight
Breakthrough
First-ever gold-level performance by general-purpose AI at IMO.
Speed and Accessibility
Naturally using human language, not coded formulas.
Potential
Near-term applications in scientific research and complex problem-solving.
Balance
Companies respected the human contestants’ recognition by timing result releases appropriately (Reuters, Axios).
What’s Next
Continued refinement of AI reasoning capabilities and integration into research workflows.
Controlled public release of these advanced models, emphasizing safety and ethical use.
Potential surge in AI-powered tools for fields like physics, mathematics, engineering—and more.