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

Starting Friday, OpenAI will stop giving opportunities to five versions of ChatGPT, including the popular but complex version of GPT-4o.

The 4o model has been among several cases related to user self-harm, fraudulent behavior, and AI psychosis. It still belongs to OpenAI the ultimate example of sycophancy.

In addition to GPT-4o, versions of GPT-5, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini have also been removed.

OpenAI plans to abandon GPT-4o in August, as it revealed Image of GPT-5. But at the time, there was enough backlash for OpenAI to make the legacy version available to paid subscribers, who could manually opt-in to the model. In a recent blog post, OpenAI noted that only 0.1% of customers have been using GPT-4o, but for a company that has 800 million users every weekthat small number is still 800,000 people.

Thousands of users have he struggled with breathing for 4ociting their intimate relationships as an example.