Sam Altman addresses ‘bumpy’ GPT-5 rollout, bringing 4o back, and the ‘chart crime’

In a recent Reddit AMA, Sam Altman, CEO of OpenAI, along with key members of the GPT-5 development team, faced a barrage of questions regarding the newly launched model. Many users expressed their desire for the return of the previous version, GPT-4o, while also bringing attention to a comical blunder known as the "chart crime" that occurred during the model's unveiling. One of the exciting features of GPT-5 is its real-time routing system, which determines the most appropriate model to respond to a user's prompt. This system allows for quicker replies or more thoughtful, extended responses. However, numerous participants in the AMA reported that GPT-5 felt less effective than its predecessor, GPT-4o. Altman addressed these concerns, explaining that the perceived drop in performance was due to issues with the router at launch. He stated, "GPT-5 will seem smarter starting today. Yesterday, we had a sev and the autoswitcher was out of commission for a chunk of the day, which made GPT-5 appear less capable." To improve the experience, Altman assured users that adjustments would be made to enhance the decision-making process of the model, ensuring it selects the right version more consistently. Additionally, he mentioned that efforts are underway to provide better transparency regarding which model is answering queries. Responding to the overwhelming requests for the return of GPT-4o for Plus subscribers, Altman confirmed that they are exploring options to allow continued access to the previous model. He also committed to doubling the rate limits for Plus users during the rollout phase, enabling them to experiment more freely with GPT-5 without the fear of exhausting their monthly usage. In light of the infamous chart error presented during the live launch—which featured misleading statistics—Altman refrained from diving into specifics during the AMA but previously admitted to the mistake on social media, labeling it a "mega chart screwup." Although the charts in the official blog post were accurate, the initial blunder sparked widespread jokes about the reliability of GPT for generating presentation visuals. Despite the hiccups, Altman concluded the session by reiterating OpenAI's dedication to stability and user feedback, promising that they would continue to refine the model as they gather more insights from users.