OpenAI has reinstated its GPT-4o model for paid ChatGPT subscribers, just one day after replacing it with the newly launched GPT-5. The reversal comes in response to widespread criticism from users who had integrated GPT-4o into their daily workflows and were unhappy with its sudden removal.

GPT-5, described by OpenAI as its most advanced large language model to date, debuted with promises of significant improvements in writing and coding capabilities. However, early reactions were mixed. During a Reddit AMA session, several users told CEO Sam Altman that GPT-5 felt “dumber” and lacked the warmer, more personable tone of GPT-4o. One user called the difference “jarring,” saying GPT-4o “genuinely talked to me,” whereas GPT-5’s replies felt “cut-and-dry corporate.”

Criticism also targeted the abrupt removal of legacy models and the elimination of the model picker feature, which previously allowed users to manually select versions like GPT-4o, GPT-4.1, or o4 mini. The company’s new real-time router—intended to automatically choose the most suitable model for each task—was another point of contention, with users claiming it was not performing as intended.

Altman acknowledged the issues, admitting that the router had been underperforming. He assured users that GPT-5 would “seem smarter starting today” and pledged greater transparency in indicating which model is responding to queries.

As of now, GPT-4o is once again available to ChatGPT Pro, Team, and Enterprise subscribers via the “show legacy models” setting. Free-tier users remain limited to GPT-5. Altman said OpenAI will monitor usage data before deciding how long to maintain support for older models, and indicated GPT-4.1 could return if demand from Plus subscribers is strong enough.

“We will continue to work to get things stable,” Altman stated, emphasizing that OpenAI will keep listening to user feedback throughout the rollout process.