Google DeepMind's Gemini 2.5 AI wins gold, stumps human coders

Google DeepMind claims its Gemini 2.5 AI has achieved a 'historic' breakthrough by becoming the first AI to win a gold medal at an international programming competition. At the ICPC held in Azerbaijan, a specially trained version of Gemini 2.5 solved a complex real-world problem involving the optimal distribution of liquid through a network of ducts to interconnected reservoirs – a task requiring deep abstract reasoning and creativity. No human team, including top performers from Russia, China, and Japan, managed to solve it. The AI completed the task in under 30 minutes and overall ranked second out of 139 of the world's strongest college-level programmers, earning a gold medal. Google likened the achievement to Deep Blue's chess victory and AlphaGo's Go triumph, calling it a 'profound leap in abstract problem-solving' and a step toward artificial general intelligence (AGI).

However, experts have tempered the excitement. The version used in the competition was not the same as the lightweight Gemini 2.5 available to subscribers of Google's $250/month AI Ultra service; Google declined to specify the exact compute power but confirmed it was more than what average users can access. UC Berkeley professor Stuart Russell called the claims 'overblown,' noting that AI systems have been improving on programming tasks for years and that the Deep Blue breakthrough had little real-world impact. Oxford professor Michael Wooldridge acknowledged the achievement but questioned the compute resources required. The AI failed two of the 12 tasks it was set, and its performance ranked second, not first. Still, ICPC executive director Dr. Bill Poucher called it a 'key moment' for defining AI tools and academic standards. The advance could eventually transform fields like drug discovery and chip design, but for now, it remains a specialized demonstration rather than a consumer-ready capability.