GPT-5 beats top teams at the Programmer’s World Championship — a new milestone for AI coding
OpenAI reports that its GPT-5 system outperformed the best human teams at the final level of the International Collegiate Programming Contest (ICPC) 2025, solving all twelve tasks — surpassing past human achievements. Human winners completed “only” eleven.
What stands out is the model’s approach: for eleven of the twelve problems, GPT-5 produced the correct answer on the first attempt. For the most difficult one, the system required nine attempts before the jury accepted its solution. This demonstrates that the model is not only capable of handling “routine” problems but also ready to tackle challenges requiring strategy and structured reasoning.
Under contest rules, GPT-5 competed without “special treatment”: the same conditions as student programmers. This result excites not only AI enthusiasts but also the industry: while coding has often remained a human stronghold, such milestones show that the gap between “scientific demo” and “practical tool” is narrowing fast.
OpenAI’s competitor, Google DeepMind, also showed strong results with its Gemini 2.5 Deep Think model, solving 10 out of 12 tasks. This underscores that in the AI race, it is no longer just about “who knows more,” but about “who is faster, more accurate, and more reliable.”
