‘suddenly Deprecating Old Models’ Users Depended On A ‘mistake,’ Admits Openai’saltman

OpenAI has brought back GPT-4o after a weekend of user protests – mostly about removal of model choice – following the rollout of GPT-5.

GPT-5 made its debut at the end of last week, with OpenAI making much of its improvements and claiming the update would cut down on hallucinations. Rather than one model, GPT-5 is a collection of models to which prompts would be routed based on factors including intent and complexity.

Simple? Not quite. Because GPT-5 could take care of the pesky decision over which model to use, OpenAI removed the menu that allowed users to choose older models, such as GPT-4o. The result was an outpouring of grief from users, many of whom had come to depend on the older models in their workflows. Each model has its strengths and weaknesses, which customers have become accustomed to.

Switching to one model “to rule them all” resulted in a cacophony of complaints and, surprisingly for a tech company, a U-turn from OpenAI.

Open AI CEO Sam Altman, perhaps tired from firefighting over the weekend, eventually confirmed in a direct answer to a user asking if 4o might be coming back: “it’s back! go to settings and pick ‘show legacy models’.”

Altman later acknowledged the disgruntlement of users and subsequent furor: “If you have been following the GPT-5 rollout, one thing you might be noticing is how much of an attachment some people have to specific AI models.

“It feels different and stronger than the kinds of attachment people have had to previous kinds of technology (and so suddenly deprecating old models that users depended on in their workflows was a mistake).”

“Mistake” is certainly one way of putting it.

Another user asked, “Am I going to be marked as weird simply if I have Legacy Models switched on? Or only if I actually use 4o a lot?

“I’m not weird I swear I just like the option.”

Altman replied, “Not marked as weird in either case!”

OpenAI also confirmed that, “by popular demand,” it is now possible to check which model was used to run a prompt.

The episode is an example of users flagging a perceived problem (some voting with their subscriptions) and a tech company appearing to pay attention and change course.

That said, OpenAI has had more than its share of problems over the years. April’s incident, in which an update to GPT-4o turned the chatbot into a sycophantic yes-bot, was rapidly rolled back following user outcry. ®

Original Source