Recently, Sheryl Hsu, a researcher at OpenAI, announced on social media that an AI reasoning system developed internally by OpenAI has won a gold medal in the 2025 International Olympiad in Informatics (IOI), marking the first time an AI has reached such a high level in this field. In a fierce competition against 330 top human competitors, the AI system outperformed 98% of the participants, with only five human competitors scoring higher than it.
Sheryl Hsu emphasized that this AI system competed under strict simulated conditions resembling those of human competitors. To ensure fairness, the AI was completely offline, unable to access the internet or external knowledge bases, and all problem-solving logic had to be generated by itself. This means its problem-solving process relied entirely on its own logical reasoning and problem-solving capabilities.
Like human competitors, this AI system also faced certain limitations, including a 5-hour competition time and 50 opportunities for code submissions. Additionally, the AI could only use a basic terminal tool, ensuring a fair showdown with human competitors in a closed and high-pressure environment.
OpenAI stated that this AI system was not specifically trained for the IOI competition, but rather used an integrated system based on a general reasoning model. During the competition, the team's human intervention mainly involved selecting the optimal solution from multiple solutions generated by the model, as well as connecting the model to the IOI submission API. Apart from these, all steps in the AI's problem-solving process were completed autonomously.
Looking back at last year's IOI competition, OpenAI had attempted to participate but failed to achieve ideal results, ultimately missing out on a bronze medal and finishing in the 49th percentile. However, just one year later, OpenAI made a significant leap from the 49th percentile to the 98th percentile, successfully winning the gold medal.
In recent times, OpenAI's internal reasoning system has achieved remarkable results in multiple world-class intellectual competitions: it ranked second in the AtCoder World Finals and claimed to have won a gold medal in the 2025 International Mathematical Olympiad. These achievements have drawn more attention to the system and sparked anticipation for the new models that OpenAI may release in the second half of the year.
Key Points:
🌟 The AI reasoning system won a gold medal in the 2025 IOI competition, outperforming 98% of human competitors.
💻 During the competition, the AI was completely offline, generating all problem-solving logic on its own to ensure fairness.
🏆 Last year it didn't win, but this year it made a qualitative leap, jumping from the 49th percentile to the gold medal.