OpenAI, the developer of ChatGPT, achieved a gold medal-level result at the International Olympiad in Informatics (IOI) 2025, which is recognized as one of the world's top programming competitions.
According to foreign media on the 13th, OpenAI's artificial intelligence (AI) system ranked 6th among human participants and 1st among AI participants in this competition. Notably, it did not undergo specialized programming training for the event, relying solely on a combination of general-purpose reasoning models.
The IOI is the premier algorithm and programming competition held annually for high school students worldwide. Participants must solve a total of six high-difficulty problems over two days, with three problems each day, under limited time and memory conditions. All problems must be written in C++ and no internet or external resources can be used.
OpenAI officially participated in the IOI's online institutional sector, subject to the same regulations as human participants. The number of submissions, time limits, and problem-solving environment were all aligned. Cheryl Shi, a researcher at OpenAI, noted, "In this competition, the model was not specially trained; we simply created a minimal structure to choose which solutions to submit and to connect to the IOI API for consolidation."
OpenAI participated in the IOI last year as well, but its performance at that time was at the bronze medal level with a correctness rate in the top 49%. However, this year it achieved an impressive 98% correctness rate, earning a gold medal. The significant improvement has garnered industry attention. In this competition, it submitted a total of 50 answers, identical to human participants, without using internet searches or retrieval-augmented generation (RAG) capabilities.
The foundation of this achievement is a general-purpose reasoning model technology known as a "universal verifier." This technology demonstrates advanced problem-solving abilities across various fields without requiring specialized training for specific problem-solving tasks. Sam Altman, CEO of OpenAI, announced plans to expand this technology to the forthcoming GPT-5, stating, "For several months, GPT-5 may not exhibit gold medal-level performance."
Meanwhile, the performance of other AI models in this IOI competition has not been disclosed. In this event, South Korea had all four participants included in the top 20, with the highest ranking reaching 2nd place.