Earlier this week, OpenAI announced GPT-5, its most powerful model to date. The rollout of GPT-5 to ChatGPT users drew a lot of criticism, as OpenAI immediately replaced all existing models, including the popular GPT-4o and o3, with the new GPT-5. ChatGPT Plus subscribers are also unhappy with GPT-5 due to its low usage limits. ChatGPT Plus plan users can send only up to 80 messages every three hours. After reaching this limit, ChatGPT switches to GPT-5 mini until the limit resets. Previously, ChatGPT Plus users had unlimited access to the o3 and o4 mini thinking models. Based on user feedback over the past two days, OpenAI is making the following changes to ChatGPT:

• The rate limits for Plus and Team plans have been doubled for the weekend.
• Next week, OpenAI will roll out mini versions of GPT-5 and GPT-5 Thinking to take over until the limits reset.
• GPT-5 Thinking and GPT-5 Pro options are now available in the main model picker.
• The GPT-4o model is also available to Plus and Team users. Users can go to the ChatGPT web interface and select the 'Show legacy models' option to enable GPT-4o.

Sam Altman confirmed on Reddit that they will monitor GPT-4o usage to determine how long to support it. During a Reddit AMA session yesterday, Sam Altman and other key OpenAI employees responded to several user questions. Here are some highlights from the AMA session:

• OpenAI may introduce another paid plan that would fall between ChatGPT Plus, which costs $20 per month, and ChatGPT Pro, which costs $200 per month.

• When the GPT-5 rollout began yesterday, the model autoswitcher experienced issues, leading to poor responses for some user queries on ChatGPT.

• OpenAI has not seen significant demand for long context lengths from its users but is open to supporting it if there is sufficient interest. Since they are compute-constrained, they want to focus on other priorities. Michelle Pokrass from the OpenAI team wrote that they would have liked to offer longer context-up to 1 million tokens-in GPT-5, particularly for API use cases, but due to high GPU demand, they did not.