In a groundbreaking achievement, an advanced version of Gemini 2.5 Deep Think has reached gold-medal level performance at the 2025 International Collegiate Programming Contest (ICPC) World Finals. This milestone follows its recent gold-medal win at the International Mathematical Olympiad, showcasing a profound leap in abstract problem-solving capabilities. The model’s success highlights how AI can tackle complex tasks requiring deep reasoning, creativity, and ingenuity—key steps toward advancing artificial general intelligence.
The ICPC is globally recognized as the most prestigious algorithmic programming competition at the college level, attracting participants from nearly 3,000 universities across 103 countries. This year’s finals in Baku, Azerbaijan, featured 139 elite teams competing under strict rules: only perfect solutions earned points, and every minute counted. Out of all competitors, only the top four teams secured gold medals.
Gemini 2.5 Deep Think competed remotely under official ICPC supervision, starting just 10 minutes after human contestants. It impressively solved 10 out of 12 problems within the five-hour time limit, achieving a performance that would rank second overall among university teams. The model solved eight problems in the first 45 minutes and utilized advanced data structures and algorithms to maintain its pace.
In an unprecedented feat, Gemini solved Problem C—a challenge that stumped all university teams—within the first half hour. This problem involved optimizing liquid distribution through a network of ducts to fill reservoirs as quickly as possible. Gemini’s innovative approach introduced “priority values” for each reservoir, applied the minimax theorem, and used nested ternary searches to efficiently navigate the infinite solution space.
This performance stems from a series of technical advances in pretraining, reinforcement learning, multi-step reasoning, and parallel thinking. Multiple Gemini agents collaborated to propose solutions, execute code, and iterate based on results. Internal studies indicate that similar versions of the model could have achieved gold-medal levels in past ICPC competitions, matching the world’s top competitive coders.
Dr. Bill Poucher, ICPC Global Executive Director, remarked: “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.”
Beyond coding competitions, this breakthrough demonstrates Gemini’s potential as a collaborative problem-solving partner. Combining the best AI and human solutions could have solved all 12 problems perfectly. These abstract reasoning skills are transferable to fields like drug design and microchip engineering, where AI can help address humanity’s most complex challenges.
Gemini users with Google AI Ultra subscriptions can already access a lightweight version of Gemini 2.5 Deep Think through the Gemini app. As AI coding assistants evolve, they promise to help developers tackle increasingly difficult engineering problems, from logistics to scientific research, ushering in a new era of human-AI collaboration.