At the International Collegiate Programming Contest (ICPC) World Finals held earlier this month in Azerbaijan, artificial intelligence models from OpenAI and Google’s DeepMind achieved gold-medal level results, successfully solving problems that human contestants could not complete. DeepMind called this a “historic moment towards Artificial General Intelligence (AGI).” The ICPC World Finals brought together 139 top teams from nearly 3,000 universities across 103 countries. Out of a total of 12 problems, OpenAI ranked first, successfully solving all 12 problems. The Gemini 2.5 Deep Think model ultimately solved 10 problems, reaching the gold medal level and securing second place.
OpenAI’s competition system consisted of multiple general-purpose models, including GPT-5 and an experimental reasoning model. Among them, GPT-5 solved 11 problems, while the experimental reasoning model solved the most difficult 12th problem and was responsible for the final submission decisions. None of the models were specifically trained for the ICPC. The OpenAI technical team stated after the competition, “I remember when I first joined OpenAI, our best AI model couldn’t even solve simple problems on Codeforces. Seeing the pace of progress over the past two years is truly incredible.”
Google introduced that the participating Gemini 2.5 was not specifically trained for the ICPC but is the version users can directly access in the Gemini application. The model was launched 10 minutes after the competition started, solving 8 problems within the first 45 minutes and completing two more within three hours. DeepMind stated that this performance represents a “leap in artificial intelligence’s ability to solve abstract problems.” For example, Gemini solved “Problem C” – a complex optimization problem concerning liquid distribution in a pipeline network to reservoirs – in less than half an hour, a problem that no human team successfully answered.
“For me, this is a moment equivalent to chess’s ‘Deep Blue’ and Go’s AlphaGo,” further explained Quoc Le, Vice President of DeepMind. “More importantly, it’s closer to the real world, not just confined environments like chess and Go. Precisely because of this, I believe this advancement has the potential to transform many scientific and engineering disciplines,” such as designing new drugs or chips. According to the latest news, Google stated that Gemini 2.5 Deep Think is a general-purpose AI model specifically trained to solve very difficult coding, math, and reasoning problems, performing “as well as the world’s top 20 programmers.” Solving complex tasks in these competitions requires deep abstract reasoning, creativity, the ability to synthesize new solutions, and a genuine spark of creativity.
Many netizens celebrated this historic moment on social media platforms. AI winning a gold medal at the ICPC might mean that artificial intelligence can become a genuine problem-solving partner for programmers. However, it has also raised concerns that human programming capabilities might forever lag behind AI in the future.
Before the details were announced, Professor Stuart Russell of Computer Science at the University of California, Berkeley, commented that “claims of being epoch-making seem somewhat exaggerated.” He believes that AI systems have been performing excellently on programming tasks for some time, and that “AI companies face immense pressure to claim breakthroughs.” He also acknowledged, however, that “to correctly answer ICPC problems, the code actually has to work correctly (at least on a limited number of test cases), so this performance might indicate that AI coding systems have made progress in generating high-quality code accurately enough.”
