OpenAI Brings GPT-4o AI Model Back to ChatGPT after User Complaints, Revises GPT-5 Thinking Rate Limits

OpenAI is finally bringing back the GPT-4o artificial intelligence (AI) model to ChatGPT, after retiring it last week. The San Francisco-based AI firm removed the older model after the launch of the GPT-5 model, highlighting that the new frontier model can handle both reasoning and conversational tasks. However, many users took to social media platforms to highlight that they preferred 4o's responses more than GPT-5's. On Tuesday, the company highlighted that the older model is now available for all paid subscribers.

Sam Altman Announces Several Changes to ChatGPT

In a post on X (formerly known as Twitter), OpenAI CEO Sam Altman listed several new updates to ChatGPT. Most of these apply to the paid subscribers — ChatGPT Plus, Team, and Pro users. The biggest announcement is the reinstatement of the GPT-4o AI model. While the model is available indefinitely, the company stated that it will provide prior notice before deprecating it in the future.

Additionally, OpenAI is also bringing back other older models, including o3, GPT-4.1, and GPT-4.5, to paid users. The 4.5 model will only be available to the Pro users, however, as it has a high computational cost. These models can be activated by visiting ChatGPT's web settings. The settings will also allow paid users to activate GPT-5 Thinking Mini.

OpenAI is also revising the weekly rate limits of GPT-5 Thinking, the advanced version of the frontier model that specialises in reasoning. Available only to paid users, this specific model will now offer 3,000 messages per week. After that, user queries will be redirected to the Thinking mini model. Notably, GPT-5 Think has a context window of 1,96,000 tokens.

Beyond this, GPT-5 is also being updated. Altman highlighted that users will now get three modes — Auto, Fast, and Thinking — when using the model. Users can manually select these options to tailor the chatbot's responses. The Auto mode will continue to intelligently route the queries to the relevant version of the model.

Altman also said that after receiving complaints about GPT-5's cold response style, the company is now working on making its personality warmer than before. He also highlighted that this update might address the immediate needs, but it is not a long-term solution. “One learning for us from the past few days is we really just need to get to a world with more per-user customisation of model personality,” he added.