OpenAI's AI model achieves "gold medal" in Math Olympiad: All the details

OpenAI

OpenAI has revealed that one of its unreleased models has achieved gold medal-level performance at the prestigious International Math Olympiad (IMO)—a rare accomplishment even among top-tier human mathematicians. Announced by OpenAI research scientist Alexander Wei on X, the AI scored 35 out of 42 points by solving five of the competition’s six notoriously tough problems.

The IMO is widely recognised as one of the world’s most challenging academic contests, testing participants on advanced algebra, geometry, and pre-calculus with problems designed to stretch human creativity and logical reasoning. That’s precisely why OpenAI’s success matters. While AI has long been proficient at processing large datasets and executing repetitive calculations, the IMO problems demand original thinking—a domain where machines traditionally struggle.

What sets this achievement apart is the model’s demonstrated ability to construct detailed, logically sound proofs, rivaling the approach of human mathematicians. Wei described it as capable of generating “watertight arguments,” a far cry from the pattern-matching shortcuts typical of most AI.

Meanwhile, OpenAI CEO Sam Altman has indicated that while this model marks clear progress, it won’t be available to the public just yet. This means that the upcoming GPT-5—while likely to show marked improvements—won’t have this full IMO-solving power baked in.

Only around 10% of this year’s 630 Olympiad participants earned gold medals, making the AI’s performance a serious flex in reasoning, not just number crunching. While it’s not clear when this level of mathematical mastery will become widely accessible, the message is unmistakable: AI is no longer just good at solving problems—it’s learning how to think through them.