Google DeepMind says a version of Gemini 2.5 achieved a “historic” milestone in problem solving. The model earned a gold medal at an international programming contest in Azerbaijan earlier this month.
According to the Guardian, it became the first AI to win gold at the event. It also placed second overall out of 139 of the world’s strongest college-level programmers.
AI solves task that beat human teams
DeepMind says Gemini 2.5 solved a complex real-world task in under 30 minutes. The challenge asked teams to choose from an infinite set of options to route a liquid through ducts to linked reservoirs. The goal was the fastest possible distribution.
No human team solved that task, including top groups from universities in Russia, China and Japan. The model failed two of 12 tasks, but its total score still ranked it second.
DeepMind called the result a “profound leap in abstract problem-solving”. The company said the system performed “as well as a top 20 coder in the world”.
Special training and access questions
The model is general purpose, but it had special training for hard coding, maths and reasoning. Google confirmed the competition system was not the same as the version in its $250-a-month Google AI Ultra service.
Quoc Le, Google DeepMind’s vice-president, said the moment is like Deep Blue for Chess and AlphaGo for Go. He said it points toward real-world reasoning, and cited drug and chip design as examples.
Scholars weigh progress and hype
Stuart Russell of the University of California at Berkeley said the “claims of epochal significance seem overblown”. He noted AI has done well on programming tasks for some time, and said the Deep Blue breakthrough had “essentially no impact on the real world of applied AI”.
He added that solving an International Collegiate Programming Contest question means the code works on test cases. That may show progress toward accurate AI coding systems. He also said “the pressure on AI companies to keep claiming breakthroughs is enormous”.
Michael Wooldridge of the University of Oxford called it impressive and exciting. He asked how much computing power it used. Google declined to say, beyond confirming it exceeded what average subscribers can access.
ICPC executive director Dr Bill Poucher said Gemini reaching gold level marks a key moment. He said it helps define AI tools and academic standards for the next generation.
