OpenAI has brought GPT-4o back to ChatGPT after users complained about the large language model’s removal.

OpenAI Chief Executive Officer Sam Altman announced the move in a Tuesday post on X. He also detailed several other changes to ChatGPT, including an increased rate limit on the newest reasoning model included in the service.

Last Thursday, OpenAI introduced a new flagship artificial intelligence system called GPT-5. It includes a reasoning-optimized LLM and a less capable model that trades off some output quality for lower latency. A component known as a router analyzes each user query and automatically determines which of the two models should process it.

OpenAI started rolling out GPT-5 to ChatGPT shortly after the AI system’s introduction last week. As part of the update, the company removed several of the chatbot service’s earlier LLMs. The move reportedly drew criticism from thousands of users, some of whom complained that GPT-5 provides lower output quality than its predecessor.

“4o is back in the model picker for all paid users by default,” Altman wrote on X. “If we ever do deprecate it, we will give plenty of notice. Paid users also now have a ‘Show additional models’ toggle in ChatGPT web settings which will add models like o3, 4.1, and GPT-5 Thinking mini.”

OpenAI introduced both o3 and GPT-4.1 this past April. The former algorithm is a reasoning model that makes about 20% fewer mistakes than o1, OpenAI’s first reasoning model, when performing “difficult, real-world tasks.” GPT-4.1, meanwhile, is a general-purpose LLM that generates most prompt responses in under a second.

Users of ChatGPT’s top-end Pro plan will have access to a third model called GPT-4.5. The LLM debuted in February and provides better performance than o1 across certain tasks. “4.5 is only available to Pro users—it costs a lot of GPUs,” Altman explained.

OpenAI has also upgraded GPT-5 as part of the update. The rate limit of GPT-5 Thinking, the reasoning model built into the newly launched AI system, is now 3,000 messages per week. Any additional queries that users send will be processed by a scaled-down version of the model with more limited reasoning features.

When GPT-5 debuted last week, OpenAI revealed plans to combine the system’s reasoning and general-purpose models into a single LLM. The planned LLM is also set to incorporate the router module that automatically determines which model should process a given prompt. It’s unclear how the complaints that GPT-5 has drawn from some ChatGPT customers will affect OpenAI’s development roadmap, if at all.

ChatGPT reportedly has around 700 million weekly users, up from 200 million last August. It’s believed that 5 million of those users are subscribed to the chatbot’s paid tiers. OpenAI expects its annualized recurring revenue to top $20 billion by year’s end. 

Photo: Focal Foto/Flickr