OpenAI's experimental AI model has demonstrated significant progress in mathematical reasoning by achieving gold medal-level performance at the 2025 International Mathematical Olympiad (IMO). Under conditions identical to those of human participants—two 4.5-hour sessions without access to external tools or the internet—the model solved five out of six problems, earning a total of 35 out of 42 points. This performance aligns with the top 10% of human contestants who typically receive gold medals in this prestigious competition.
The IMO is renowned for its challenging problems that require creative and rigorous logical reasoning. OpenAI's model was evaluated under the same conditions as human participants, including two 4.5-hour sessions, no access to tools or the internet, and writing detailed proofs based on official IMO problems. The AI successfully solved five out of six problems, earning 35 out of 42 possible points. Three former IMO medalists graded each solution independently, with final scores based on unanimous agreement.
This achievement underscores the rapid advancements in AI's reasoning capabilities. OpenAI's approach involved developing a general-purpose reasoning language model that can craft intricate, watertight arguments at the level of human mathematicians. The model's success at the IMO demonstrates progress in general-purpose reinforcement learning and test-time compute scaling.
While this experimental model showcases impressive capabilities, OpenAI does not plan to release anything with this level of math capability for several months. The upcoming GPT-5, which is expected to be released soon, will likely be an improvement from its predecessor but won't feature the same level of mathematical prowess as the IMO-winning model.
This milestone highlights the potential for AI to assist in education and research, making complex concepts more accessible and engaging for students and professionals worldwide. The implications are vast, suggesting a shift towards more interactive and personalized learning experiences, impacting how we teach and learn.