OpenAI restores GPT-4o after user backlash; Sam Altman promises fixes

In a remarkable and swift reversal, OpenAI has reinstated access to its popular GPT-4o model for ChatGPT Plus subscribers, responding to a groundswell of user dissatisfaction and a wave of subscription cancellations that followed the abrupt rollout of its latest model, GPT-5. The decision, confirmed by CEO Sam Altman, underscores the critical importance of user experience and emotional connection in the rapidly evolving landscape of artificial intelligence.

Just days ago, OpenAI launched GPT-5, touting it as its most advanced system and making it the default model for all ChatGPT users, including paid subscribers. Crucially, this update removed the option for users to select older models, automatically defaulting to GPT-5 and employing a “real-time router” to switch between different variants of the new model based on the task. However, the intended seamless transition quickly spiraled into widespread user backlash across social media platforms and online forums like Reddit.

Users swiftly voiced their frustration, describing GPT-5’s responses as “sterile,” “brief,” “colder,” and lacking the “personality” or “emotional intelligence” that had become hallmarks of GPT-4o. Many reported a decline in conversational style and perceived writing quality, finding the new model less engaging and more mechanical. For some, the shift was deeply personal; they had formed genuine “emotional connections” with GPT-4o, viewing it as a “trusted friend” or a “soul companion” that complemented their workflows and communication styles. The forced upgrade also disrupted carefully built routines, with several users lamenting that it broke their established workflows.

Adding to the discontent were concerns over restrictive message limits imposed on Plus users with GPT-5, which appeared significantly lower than previous allowances for GPT-4o and other models. The combination of perceived performance degradation, a colder tone, and reduced flexibility led to a notable number of users canceling their ChatGPT Plus subscriptions in protest, perceiving GPT-5 as a “downgrade” rather than an advancement. Further compounding the issue, Altman admitted that technical glitches with the “autoswitcher” during the rollout had caused GPT-5 to “seem way dumber” at times.

In response to the unprecedented user revolt, Sam Altman took to Reddit and X (formerly Twitter) to address the concerns directly. He acknowledged that OpenAI had “underestimated how much some of the things that people like in GPT-4o matter to them,” even if GPT-5 performed better in most benchmarks. As a direct concession, Altman confirmed that GPT-4o would be restored as a selectable option for Plus subscribers, accessible via a “show legacy models” setting.

Beyond the reinstatement, OpenAI has pledged a series of improvements. The company plans to double GPT-5’s rate limits for Plus users and is testing a significantly higher cap of 3,000 reasoning requests per week for the GPT-5 Thinking model. In the near term, efforts are underway to make GPT-5’s tone “warmer,” while longer-term investments will focus on “steerability,” allowing users greater customization over the system’s tone and emotional intelligence. OpenAI also promised clearer model labeling in the interface and easier ways to manually trigger “thinking mode,” alongside enhanced transparency about which model is responding to a query.

While the immediate return of GPT-4o addresses pressing user concerns and aims to rebuild trust, the long-term availability of older models will depend on usage patterns. This episode serves as a stark reminder for AI developers that technical superiority alone may not guarantee user adoption, and that the nuanced, often emotional, connection users form with AI tools is a vital factor in their success.