In the 2025 International Collegiate Programming Contest (ICPC) World Finals, OpenAI's system performed outstandingly, successfully solving all 12 programming problems in the competition, surpassing human contestants and Google's DeepMind model. According to OpenAI, if this system had participated as a human team, it would undoubtedly have won first place.

Code Internet Computer

Image source note: The image was generated by AI, and the image licensing service is Midjourney

OpenAI stated that the system followed the same rules as students during the competition. It obtained the problem set in standard PDF format and submitted solutions within a five-hour time limit. All submissions were directly sent to the official judges, who evaluated them alongside the solutions from human teams. The company emphasized that no special adjustments were made to the system during the competition.

Shortly after this news was released, Google's DeepMind claimed that its upgraded Gemini 2.5 Deep Think system achieved gold medal level in the same competition, successfully solving 10 out of 12 problems, including problem C, which human teams failed to solve. However, Gemini failed on two other problems. In comparison, OpenAI's system achieved a perfect score of 12/12, surpassing DeepMind's performance.

OpenAI explained that the system is an ensemble of multiple general reasoning models, which were not specifically trained for ICPC. It used GPT-5 and an internal experimental reasoning model. GPT-5 provided correct answers for 11 out of 12 problems, while the experimental model was responsible for selecting the final answer to submit. The last and most difficult problem was solved independently by the experimental model, which required nine submissions before being finally accepted. In contrast, the best-performing human team solved only 11 problems.

OpenAI emphasized that this victory demonstrates the potential of more complex models in deep reasoning and long-term computational capabilities. GPT-5 handled most of the work, highlighting its status as one of the most powerful AI models currently available to the public. The company views this achievement as a result of continuous progress. These models have already achieved gold-level results in important academic competitions such as the International Mathematical Olympiad and the International Olympiad in Informatics. Mustafa Rohani Jadid, a project participant, stated that the performance in ICPC is a fitting conclusion to this series of successes and noted the system's versatility. He also added that the next step will be to develop systems capable of discovering new knowledge, which is considered a real milestone.

Key Points:

- 🏆 OpenAI's system perfectly solved all 12 problems in the 2025 ICPC World Finals, surpassing humans and Google's DeepMind.

- 🤖 The system combines GPT-5 and an experimental model, showcasing its advantages in complex reasoning tasks.

- 📈 OpenAI sees this victory as a sign of improvement in AI capabilities, and will continue to explore the discovery of new knowledge in the future.