OpenAI began notifying users on November 10 that Advanced and Standard voice will retire from ChatGPT’s voice model picker on December 11. People who still have either option selected will be moved to Live voice automatically, according to the company’s retirement notice.
The transition does not delete earlier conversations. OpenAI says those chats will remain in history and can continue in text or voice. Users can keep choosing the older modes during the transition period, giving them time to compare the experience before the December deadline.
A separate timetable for video
There is an important exception to the retirement. Live video and screen sharing will continue to use Advanced voice until GPT-Live supports those capabilities. OpenAI has not given a date for that replacement in its notice, so the removal of a general voice choice should not be read as the simultaneous end of those visual features.
What Live currently supports
OpenAI’s current ChatGPT Voice help page describes Live as an experience that works within a chat, with spoken responses also available as text. It can use web search, memory, supported visual results and available connected apps. Access depends on factors such as plan, workspace settings, region, app version and parental controls.
The help page explicitly says Live does not yet support video or screen sharing. It also notes that actions requiring approval on web and mobile must be reviewed through on-screen controls; spoken approval is not supported for those prompts. A natural conversation therefore does not remove the need to inspect an action before accepting it.
Voice can still make mistakes. OpenAI advises users to check important answers and be specific about dates, locations and time zones when those details matter. Changing the voice system does not turn a spoken answer into a verified source of information.
Why this is more than a different sound
In its original July 8 GPT-Live announcement, OpenAI described a full-duplex architecture, meaning the system can listen while it speaks. Earlier Standard voice linked speech recognition, a language model and speech generation. Advanced voice processed audio directly but still handled conversation as discrete turns.
GPT-Live was designed to make interaction decisions continuously, including whether to speak, listen or pause. The launch description also separated conversation from deeper work: the voice model could pass a difficult request to another model and continue talking while that work was underway. These are architectural differences, not simply new voice recordings.
For users who depend on a particular interaction style, the useful next step is to try Live on familiar tasks before December 11. The remaining uncertainty is when video and screen sharing will move to it. Until OpenAI announces that change, the company’s published guidance keeps those capabilities on Advanced voice.
