OpenAI’s new model was able to get 5 out of 6 questions (a gold medal) on the 2025 International Math Olympiad. I am very surprised by this result, though I don’t see any evidence of foul play.