OpenAI removes access to sycophancy-prone GPT-4o model

In Brief

Starting Friday, OpenAI volition cease providing entree to 5 bequest ChatGPT models, including the fashionable but arguable GPT-4o model.

The 4o exemplary has been astatine the halfway of a fig of lawsuits concerning idiosyncratic self-harm, delusional behavior, and AI psychosis. It remains OpenAI’s highest scoring exemplary for sycophancy.

In summation to GPT-4o, the GPT-5, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini models person besides been deprecated.

OpenAI intended to discontinue GPT-4o successful August, erstwhile it unveiled the GPT-5 model. But astatine the time, determination was capable backlash for OpenAI to support the bequest exemplary disposable for paid subscribers, who could manually take to interact with that model. In a caller blog post, OpenAI noted that lone 0.1% of customers person been utilizing GPT-4o, but for a institution with 800 cardinal play progressive users, that tiny percent inactive amounts to 800,000 people.

Thousands of users person rallied against the status of 4o, citing their adjacent relationships with the model.

Subscribe for the industry’s biggest tech news