OpenAI's AI system wins a gold medal-level score at the International Olympiad in Informatics 2025

Unlike previous years, the OpenAI team did not train a custom model specifically for the IOI. Instead, they used an ensemble of general-purpose logic models. OpenAI researcher Noam Brown highlighted that the top-performing core model was the same one that recently earned a gold medal at the International Mathematical Olympiad (IMO). After extensive evaluations, the team found this model excelled not just in competitive math, but also in other areas like programming, and deployed it to the IOI without modification.

This result marks a significant leap forward in just one year. In 2024, OpenAI's system narrowly missed a bronze medal, landing in the 49th percentile. At that time, the team relied on a much more handcrafted testing strategy, including a heavily fine-tuned model supported by a complex scaffold with synthetic test cases and manual features. This year, the approach was far simpler: a lightweight scaffold only selected the best solutions from the general models' outputs using another model and a basic heuristic. The streamlined process propelled the AI to the 98th percentile. Brown suggested that by next year, the model itself could outperform any scaffold the team might devise.