OpenAI has introduced a new paper detailing advancements in competitive programming using large reasoning models, with notable achievements by its o3 model. The o3 model earned a gold medal at the 2024 International Olympiad in Informatics (IOI), scoring 394 out of 600 points and ranking 18th globally. Additionally, o3 achieved a Codeforces rating of 2724 Elo, placing it in the 99.8th percentile, comparable to elite human programmers. The o1-ioi model, specifically designed for the IOI, also achieved gold-medal results. Unlike its predecessor, o1, which relied on hand-crafted pipelines, o3 demonstrated superior performance using general-purpose reinforcement learning techniques. The results adhered to the 50 submission limit, ensuring data integrity. Kevin Weil from OpenAI confirmed that o4, the next iteration, is already in training. This development highlights the potential of scaling general-purpose reinforcement learning over domain-specific strategies in competitive programming and reasoning tasks.