Google’s Gemini 2.5 Deep Think cracks problem no team could solve at global programming contest
The progress shown in both competitive programming and mathematical reasoning suggests that Gemini is advancing in its ability to deal with abstract challenges.
In a contest designed to test the limits of the world’s best student programmers, an advanced version of Gemini 2.5 Deep Think did what no university team could, by cracking a problem that remained unsolved by all human competitors.
Within the first thirty minutes, it produced a working solution to Problem C, which asked competitors to find the fastest way to distribute liquid through a network of ducts into several reservoirs. The difficulty lay in the vast number of possible configurations, since each duct could be open, closed or partly open.
The achievement set Gemini apart in a contest regarded as the most demanding arena for university-level programming. Success at the ICPC requires more than coding skill. It calls for abstract reasoning, creativity and the ability to design solutions to unfamiliar problems, often with only a small spark of insight to guide the way.
The progress shown in both competitive programming and mathematical reasoning suggests that Gemini is advancing in its ability to deal with abstract challenges. For researchers, this marks a significant step towards developing systems with more general forms of intelligence.
ICPC is the world’s longest-running and largest competition focused on algorithmic problem-solving. Each year it draws participants from close to 3,000 universities across more than 100 countries. In the 2025 finals, 139 teams were given five hours to tackle a series of demanding programming challenges.
Gemini 2.5 Deep Think took part, competing remotely under the same rules and oversight as the student teams. Despite beginning ten minutes later, it successfully solved 10 of the 12 problems within the time limit. According to the official scoring system, this would have placed it second overall, with a total time of 677 minutes.
“The ICPC has always been about setting the highest standards in problem solving. 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,” said Dr Bill Poucher, ICPC Global Executive Director.