Every year, the International Collegiate Programming Contest (ICPC) brings together thousands of the world’s best student coders. They face 12 brain-bending algorithm puzzles in just five intense hours. To make it tougher, points are awarded only for correct answers, and faster solutions mean higher ranks. For the 2025 contest, Google entered Gemini 2.5 Deep Think, plugging it into a secure online environment approved by ICPC. Human competitors were given a 10-minute head start before the AI joined in. Gemini’s performance Unlike Google’s custom-trained AI for the Math Olympiad earlier this year, Gemini wasn’t specially rebuilt for ICPC. It used the same general-purpose model available in other Gemini apps just enhanced to think continuously for hours. The results? Stunning Out-thinking humans One standout was Problem C, a multi-dimensional optimization puzzle about storing and draining imaginary “flubber.” Every human team failed it but Gemini cracked it in 30 minutes. By assigning priority values to each reservoir and applying dynamic programming with nested ternary search, the AI found the optimal configuration. Beyond one contest Gemini’s abilities weren’t just a one-off. When tested on past ICPC problems from 2023 and 2024, it again reached gold-level performance. ICPC director Bill Poucher noted: “Gemini’s success sets a new benchmark for AI in problem-solving and academic standards.” Google believes this kind of multi-step reasoning could transform industries like semiconductor design and biotechnology, where complex logic is key.
The post Google’s Gemini just beat every human team at coding contest:Sundar Pichai announces Gemini 2.5 Deep Think achieves gold-medal performance at 2025 ICPC World Finals appeared first on Tri-Cities India.