OpenAI Reverses Course: GPT-4o Returns After Fan Outcry to Sam Altman
The recent rollout of OpenAI’s latest flagship model, GPT-5, has been met with an unexpected torrent of user backlash, compelling CEO Sam Altman to announce the reinstatement of its predecessor, GPT-4o, for paying subscribers. What OpenAI intended as a significant upgrade quickly devolved into widespread frustration, with users taking to platforms like Reddit to express their dismay, some even reporting “tears” over the perceived downgrade.
The controversy ignited swiftly after OpenAI replaced previous models, including the widely favored GPT-4o, with GPT-5 without prior warning. This abrupt transition stripped users of their ability to select specific models, a feature many had grown accustomed to. The core of the dissatisfaction stemmed from a perceived decline in GPT-5’s performance and “personality” compared to GPT-4o. Users lamented that GPT-5 felt “dumber” or less capable, particularly in tasks requiring complex reasoning, coding, and even basic accuracy. Complaints flooded online forums, with many reporting shorter, less nuanced responses, and a general lack of the “warmth and understanding” that made GPT-4o feel “human” to some. Some users even compared the experience to “losing a friend overnight” or watching a “close friend die”.
Adding to the frustration were stricter usage limits imposed on ChatGPT Plus subscribers, who found their access to the new GPT-5 “Thinking” model capped at 200 messages per week, a significant reduction in functionality compared to previous offerings. This led many paid users to question the value of their subscriptions, with some threatening to cancel.
Sam Altman quickly acknowledged the “bumpy” rollout and the overwhelming user feedback during a candid Reddit AMA on August 8, 2025. He attributed some of the initial performance issues to a “malfunctioning real-time router system” (or “autoswitcher”) that was intended to seamlessly select the optimal model for each query but instead caused inconsistent and often inferior outputs. Altman stated that this system was “out of commission for a chunk of the day,” making GPT-5 seem “way dumber” than intended. He promised immediate interventions to fix this “decision boundary” and enhance transparency by making it clearer which model is answering a query.
In a direct response to the fervent pleas, Altman confirmed that OpenAI would allow ChatGPT Plus subscribers to revert to GPT-4o, at least temporarily, while the company monitors usage and considers long-term support for legacy models. He also announced that GPT-5 rate limits for ChatGPT Plus users would be doubled to ease the transition. Beyond the technical glitches, the incident highlighted the deep emotional connection many users form with AI models, particularly GPT-4o, which was praised for its intuitive voice response, multimodal capabilities, and ability to engage in natural, real-time conversations. This “vibe check,” as internal experts at OpenAI call it, proved that automated evaluations sometimes miss crucial aspects of user experience.
The rapid reversal by OpenAI underscores the critical role of user feedback in the development and deployment of advanced AI. It also serves as a potent reminder that even with groundbreaking technology, user sentiment, perceived utility, and the subtle nuances of AI “personality” can significantly impact adoption and satisfaction. As OpenAI navigates these challenges, the incident emphasizes the ongoing balance between pushing technological boundaries and ensuring a stable, user-centric experience, especially as competitors continue to advance their own AI offerings.