This marks our steady progress in math and programming competitions. Just over a year ago, we launched OpenAI o1-preview and OpenAI o1-mini. Since then, our general reasoning models have steadily improved. Today, they achieve top scores in some of the world's toughest math and programming competitions: gold at the International Mathematical and Informatics Olympiad, runner-up at the AtCoder Inspirational Programming Challenge, and most recently, a perfect score on all 12 problems at the 2025 ICPC World Finals. The ICPC World Finals is the world's premier collegiate programming competition, often referred to as the "Olympics of Programming." The finals brings together elite student teams from over 100 countries to solve 12 algorithmic problems in just 5 hours, all subject to strict time, memory, and runtime constraints. Achieving first place is crucial, as ICPC has long been the gold standard for testing human programming ability. Many of our researchers have competed in the ICPC World Finals, including @merettm and @bminaiev. We employed a simple yet powerful approach: using GPT-5 and our in-house experimental reasoning model to simultaneously generate multiple candidate solutions, and then using our experimental model to intelligently select the best submission. No complex strategies or frameworks were involved. Our model competed under the same conditions as human participants, including the same hidden test cases, time limits, memory constraints, hardware specifications, and sandbox evaluation environment. In its first submission to the ICPC-managed and sanctioned online review environment, GPT-5's solution successfully solved 11 of the 12 problems. After GPT-5 struggled, our experimental reasoning model tackled the final and most challenging problem. It took a total of nine submissions for both GPT-5 and the experimental model to solve it. This highlights how smarter models, capable of deeper reasoning and with longer computational time, can succeed on more complex tasks. GPT-5 solves most of these problems, demonstrating our achievement in providing one of the smartest and most widely used AI models to the public. Just a year ago, our inference models were far less intelligent and significantly more expensive. Now, we've developed models that are not only smarter, but also faster and more affordable, making powerful AI inference capabilities accessible to everyone with GPT-5. Finally, congratulations to all @ICPCNews participants this year. The talent, creativity, and determination you bring to the competition each year are truly remarkable.