ChatGPT sets December 11 switch from older voice modes to Live

OpenAI will remove Advanced and Standard from ChatGPT's voice model selector on December 11, it announced on November 10. Accounts that have either option selected will switch automatically to Live on the retirement date. People can continue using their existing mode during the transition period.

Live video and screen sharing have a separate handover condition. They will continue to run on Advanced voice until GPT-Live supports those capabilities. OpenAI has given no date for that handover, so the December deadline covers the selectable voice modes while visual input follows its own support schedule.

A change to how conversations flow

OpenAI says voice usage has nearly doubled since GPT-Live arrived in July and that almost all people using voice in ChatGPT now use it. The company says consolidating the experience will let it focus on GPT-Live. It identifies response depth, conversational continuity and personalization as areas for improvement. The notice gives no absolute usage total or detailed counting method.

Live uses a full-duplex architecture that processes listening and speaking at the same time. It can decide whether to speak, keep listening or pause as a person adds a thought or stops to think. OpenAI's July launch description separates that continuous interaction from deeper work. Search, reasoning and complex requests can pass to another model, with the result brought back into the spoken conversation.

Standard voice connected three models in sequence to transcribe speech, generate an answer and speak it. Advanced processed audio directly in one model, with responses organized around completed speaking turns. The move therefore changes the underlying conversational process. OpenAI updated Live on October 15 to GPT-Live-1.1 and the smaller Quip model, adding voices for languages including Korean.

Existing chats and audio retention

Conversations made with the older modes remain in ChatGPT history and can continue in text or voice. OpenAI's transition guidance says prior context, including interests, preferences and goals shared in a conversation, can carry over to newer models. The guidance also acknowledges that a model's tone and response style may change, and says feedback informs improvements to behavior and personalization.

Audio storage also differs between the modes. Live and Advanced audio clips, plus Advanced video clips, are stored with the transcript in chat history and retained for 30 days. Standard audio is deleted after transcription, except where the user has chosen to share it for model training. Keeping a conversation in history and keeping its original recording follow different rules.

Deleting a chat normally deletes its associated audio and video clips within 30 days. Exceptions cover security, safety or legal retention and shared training clips already disassociated from an account. Archiving hides a chat from the sidebar, and recordings follow the mode’s retention policy.

Dictation and a copy of chat history

ChatGPT Dictation remains a separate interaction for turning one recording into editable text. It lets the user review and change the transcribed prompt before sending the message. Dictation audio is retained while the chat stays in history. Deleting the chat normally schedules that recording for deletion within 30 days, with security, legal and previously disassociated training-data exceptions.

A copy of eligible account data can be requested through data export, available to Free, Go, Plus, Pro and qualifying Edu accounts. The ZIP includes chat history and related account data, with conversations in JSON and a viewer for searching chats and previewing attachments. The viewer processes files locally in the browser. Business, Enterprise and Healthcare data access is handled through the workspace owner. An export requested in settings can take up to seven days to arrive, and its download link expires 24 hours after receipt.