OpenAI has reinstated its GPT-4o language model for paying ChatGPT Plus customers after a wave of complaints about the company’s newly launched GPT-5. The move, announced by Chief Executive Officer Sam Altman on 11 August, reverses last week’s decision to discontinue all legacy models as GPT-5 rolled out to the platform’s hundreds of millions of users. Users said GPT-5 felt slower at complex reasoning tasks, less personable in everyday conversation and prone to coding errors after an autoswitching bug made the system appear “way dumber” than intended. The backlash underlined how strongly some customers had grown attached to GPT-4o’s warmer tone, prompting Altman to concede that “suddenly deprecating old models that users depended on was a mistake.” Under the revised plan, ChatGPT Plus subscribers, who pay about US$20 a month, can once again select GPT-4o through a new “legacy models” toggle. At the same time, OpenAI is doubling GPT-5 usage ceilings to 160 messages every three hours and raising the cap on the higher-powered “Thinking” mode to 3,000 messages a week. A forthcoming user-interface update will also show which model generates each reply to improve transparency. Altman said the company is working to fix remaining technical issues and to give users more ways to tailor the assistant’s personality and safety settings. He framed the episode as a lesson in balancing rapid innovation with the emotional and workflow dependencies that increasingly form around generative AI tools. OpenAI has not specified how long GPT-4o will remain available, saying only that it will monitor demand before deciding when to retire the model permanently.