OpenAI CEO Sam Altman has announced new updates for ChatGPT on X. The changes follow criticism from users after the launch of GPT-5, which saw older models such as GPT-4o suddenly removed.
Users will now be able to choose from three modes for GPT-5: Auto, Fast, and Think. The Auto setting automatically selects the most appropriate mode based on the question, while Fast is optimized for low-latency responses and Think is intended for more complex tasks that require more reasoning. According to Altman, the majority of users will use Auto, but the additional choices offer more control for those who want it.
The usage limits have also been adjusted. The limit for GPT-5 Think has been increased to 3,000 messages per week. Once this limit is reached, ChatGPT automatically switches to GPT-5 Think mini. The context limit for GPT-5 Think has been set at 196,000 tokens. OpenAI says these limits may be adjusted in the future depending on usage.
GPT-4o returns
At the request of the community, GPT-4o has been brought back and is now available by default in the model selector for all paying users. According to Neowin, the sudden disappearance of GPT-4o led to emotional reactions on Reddit, with some users describing the model as their best friend or even a kind of soulmate.
If the model is phased out in the future, OpenAI will announce this well in advance. In addition, a new option has been added to the ChatGPT web settings for paying users, which makes additional models such as o3, 4.1, and GPT-5 Think mini visible. The GPT-4.5 model remains reserved for Pro users due to high GPU costs.
Neowin writes that some of the negative experiences with GPT-5 were related to a problem in the automatic router, which determines when the system should switch between fast and thinking mode. In some cases, this router got stuck, causing the model to perform less well. Neowin also reports that when GPT-5 was launched, free users could only send ten messages every five hours and could only use GPT-5 Thinking once a day. Plus users had a limit of eighty messages every three hours, after which they were switched to GPT-5 mini.
Finally, OpenAI is working on an update to GPT-5’s personality. The model should feel warmer than the current version. But also less disturbing than GPT-4o. According to Altman, the events and feedback of the past few days have made it clear that OpenAI wants to focus much more on personalization options in the future, so that each user can set the tone and style of their AI as they wish.