Gemini 2.5 Deep Think achieves gold-medal level at ICPC 2025

The International Collegiate Programming Contest (ICPC) is considered the oldest and most prestigious university-level algorithmic programming contest. According to Google, the best teams from over 100 countries competed in the 2025 World Championship in Baku, Azerbaijan. Complex algorithmic problems had to be solved within five hours. Points were awarded only for error-free solutions. Of 139 teams, only four won gold medals.

Gemini solves ten of twelve problems

According to Google, Gemini 2.5 Deep Think went online ten minutes after the human teams and solved eight problems within 45 minutes and two more within three hours. A total of ten of twelve problems were solved correctly in a total time of 677 minutes, which would have placed Gemini second among the university teams. Dr. Bill Poucher, ICPC Global Executive Director, commented:

Gemini successfully joining this arena, and achieving gold-level results, marks a key moment in defining the AI tools and academic standards needed for the next generation. Congratulations to Google DeepMind; this work will help us fuel a digital renaissance for the benefit of all.

Problem C: A task that no university team solved

Particularly noteworthy is Problem C, which no human team was able to fully solve. The task was to configure a network of pipes in such a way that several reservoirs could be filled as quickly as possible. Gemini combined dynamic programming, the minimax approach and nested ternary searches to find priority values for optimal flow in order to solve the problem.

Technical methods and learning strategies

According to Google, Gemini's performance is based on advances in pretraining, post-training, novel reinforcement learning techniques, multi-step reasoning and parallel thinking. Multiple AI agents propose their own solutions, test them, execute code, and iterate until the optimal solution is achieved. According to Google, internal analyses show that Gemini would have also achieved gold level at the 2023 and 2024 ICPC World Finals.

Practical relevance for research and development

The results demonstrate that AI can act as a problem-solving partner for developers. Combining AI with human teams may help to solve large and complex tasks. This opens up potential for software development, scientific research and complex engineering projects. Google AI Ultra subscribers can already use a simplified version of Gemini 2.5 Deep Think via the Gemini app.