Historic Milestone in AI Math Reasoning
Artificial intelligence tools from
Google Gemini and
ChatGPT (developed by OpenAI) have reached a landmark achievement at the 2025 International Mathematical Olympiad (IMO), winning gold medal-level scores for the first time in the competition’s history. The IMO, held this year in Australia, is recognized as the world’s most prestigious math contest for high school students, with participation from over 100 countries.
How the AI Systems Performed
- Google DeepMind officially entered the competition with its advanced Gemini Deep Think model, solving five out of six problems perfectly and earning 35 out of 42 points—an achievement certified by IMO judges.
- OpenAI, while not an official entrant, tested its new model on the same 2025 IMO problems. The results, also totaling 35 points, were evaluated by three former IMO medalists who reviewed and approved the AI’s multipart mathematical proofs.
Both AI models abided by human contestant rules: two 4.5-hour exam sessions with no internet or external tools permitted.
Complex Problems, Advanced Solutions
The IMO questions challenge participants across high school-level algebra, combinatorics, geometry, and number theory, but demand multi-page, rigorous proofs with dense mathematical formulas and clear explanations. Only 67 contestants out of 630 (about 10%) earned gold medals this year, highlighting the elite nature of the contest and the magnitude of the AI systems’ success[1][2].
Technological Innovation Behind the Wins
- Google’s Gemini Deep Think model leverages a feature enabling it to generate and combine multiple solution paths, boosting accuracy on complex problems.
- ChatGPT’s experimental model signifies unprecedented progress in general-purpose reinforcement learning and mathematics reasoning. However, this model is not currently available to the public and will remain research-only for several months[3][4].
Debate Over OpenAI’s Gold Status
Although both models achieved matching scores, some experts have questioned the comparability, as Google DeepMind’s results were certified under official IMO procedures, while OpenAI’s grading relied on independent review by former medalists rather than by the competition’s adjudicators[4]. Concerns remain about benchmarking consistency until more transparency or public release occurs.
AI’s New Role in Advanced Mathematics
The twin gold-level results mark a breakthrough in AI’s ability to tackle sophisticated, multi-step reasoning tasks previously reserved for human intellect. As AI models continue to evolve, their performance at events like the International Mathematical Olympiad hints at a future where collaborative human-AI problem solving could become standard practice in mathematics.
