OpenAI will retire Standard and Advanced, two of the three voice modes that let people talk with ChatGPT out loud, on December 11, 2026. Anyone still using them that day will be moved automatically to the third mode, Live. Live runs on GPT-Live, the voice model OpenAI launched in July. The company announced the change on November 10 in the ChatGPT release notes, on its Help Center and on its OpenAI Newsroom account on X.
OpenAI cites usage to explain the move. It says voice use in ChatGPT has nearly doubled since GPT-Live arrived in July. It also says virtually all voice users are already on Live. Some subscribers have been pushing back on Reddit and X since the announcement.
The timeline OpenAI set
OpenAI’s Help Center page on the retirement lays out three steps. Users were notified on November 10, 2026. Until December 11, they can still choose Standard or Advanced in the Voice section of Settings. On December 11, both options disappear and the accounts that used them switch to Live.
Past voice conversations stay in the chat history. OpenAI says users can continue them in text or by voice, with Live as the default voice after December 11.
December 11 is also the day custom GPTs are retired. Custom GPTs are the assistants anyone can build in ChatGPT with their own instructions and files. Voice in custom GPTs still runs on Advanced, so it ends when the GPTs do.
Three generations of voice modes
Standard is the oldest mode. It chains three AI models: the first turns speech into text, the second writes the reply and the third reads it aloud. OpenAI acknowledges that this setup made answers slow and that some information was lost between models.
Advanced, the second generation, handles audio inside a single model, which shortens the wait. It still works turn by turn, though: it waits for the user to stop talking before it answers. A pause or background noise can make it jump in too early.
GPT-Live, the voice model behind Live, came to ChatGPT on July 8, 2026. It listens and speaks at the same time, which OpenAI calls “full duplex.” Users can interrupt it, and it can drop in an “mhmm” to show it is following along.
For a web search or a hard question, GPT-Live hands the work to a more powerful text model while the conversation keeps going. At launch, that model was GPT-5.5. Since September 9, it can be GPT-5.6 or GPT-6 Astra. GPT-6 Astra is one of OpenAI’s most powerful models, and our guide to AI models compares it with its rivals.
At first, Live was available only on ChatGPT’s consumer plans, the ones for individuals. On July 23, OpenAI opened it to Business, Enterprise and Edu workspaces, the organization-wide versions of ChatGPT for companies and schools. On August 7, Live became the default voice mode in Enterprise, Edu and Healthcare workspaces. Healthcare is the version for health organizations.
At launch, OpenAI warned that GPT-Live could have a non-native accent or gaps in fluency in some languages. On October 15, the company rolled out GPT-Live-1.1, an update that it said brings more natural accents across languages. It added new voices the same day, including two French ones, Aube and Rivage.
Video and screen sharing stay on Advanced
Advanced will stay in place for two features that Live cannot handle yet: live video and screen sharing. Both are available to eligible subscribers in the ChatGPT apps for iOS and Android. OpenAI says they will keep running on Advanced until GPT-Live supports them. The tech news site Android Authority noted that OpenAI gave no timeline for that.
Audio kept for 30 days rather than deleted
The end of Standard also changes what happens to recordings. According to OpenAI’s Help Center page on voice, Standard turns speech into text and deletes the audio as soon as the transcription is done. The recording is kept only if the user agreed to share audio with OpenAI to improve its models.
The same page says that with Live and Advanced, audio clips are stored with the conversation’s transcript and kept for 30 days. Deleting the conversation deletes the clips within 30 days. OpenAI says it does not use these clips to train its models unless the user agrees. Transcripts can be used for training when the “Improve the model for everyone” setting is on, depending on the plan and settings.
The November 10 announcement does not mention dictation, a separate feature from voice mode. Dictation is the microphone button that turns speech into text you can edit before sending.
Daily limits that vary by plan
According to the voice Help Center page, Live’s limits are counted over a rolling 24-hour window, meaning the past 24 hours at any given moment. Free accounts get limited access to GPT-Live-1.1 Quip, the lighter version of the model. Subscribers to Go and Plus, two of ChatGPT’s paid plans, get 3 hours a day, with the lighter version on Go. The Pro plan comes at two prices: the cheaper one gets 15 hours a day and the more expensive one gets unlimited use.
On ChatGPT Business, the cap depends on the type of seat, or user license. A Business Standard seat gets 3 hours a day and a Business Premium seat gets 15 hours. Past that cap, each extra minute costs 1.25 credits, the unit OpenAI uses to bill extra usage. Before September 9, the rate was 5 credits a minute, four times as much.
OpenAI’s release notes add one more point for Business and Enterprise workspaces. When Live hands a search or a reasoning task to a text model, that work is counted separately, against the text model’s usual limits. Those text-model limits also change in December for Plus, Pro and Business subscribers. ChatGPT and Codex, OpenAI’s coding tool, will then share one allowance, as our article on the shared limit explains.
A retreat in 2025, a second attempt in 2026
OpenAI has tried to retire Standard before. On August 7, 2025, its release notes said the mode would be retired in 30 days. On September 9, 2025, the company reversed course. It wrote that Standard mattered a great deal to some users and that it would keep the mode while it addressed their feedback about Advanced.
This time, OpenAI is relying on the fact that most voice users have already moved to Live. Its Help Center page acknowledges that these transitions “can be challenging” for people used to the way the older modes respond. The page links to a general article about model changes. That article points people who find the transition emotionally difficult to support resources, including the crisis support offered in ChatGPT.
Subscriber complaints and Live’s current limits
On November 11, the tech news site Android Headlines rounded up complaints posted on Reddit and X. Longtime users say GPT-Live jumps in too quickly and cuts off their train of thought. Others say familiar ChatGPT voices such as Vale sound different with GPT-Live. Android Headlines said TechRadar, another tech site, had flagged that point earlier.
The same day, the French tech site LeBigData.fr quoted a subscriber who said Live forgets context and asks questions that were already answered. On Reddit, a post published November 15 pointed readers to online petitions against the change.
OpenAI’s voice Help Center page also lists some current limits of Live. ChatGPT’s preset personalities, which set the tone of replies, do not apply to Live yet. There is no precise control over speaking speed either: users have to ask ChatGPT to talk faster or slower.
Numbers OpenAI doesn’t break down
The announcement rests on two claims: voice use has nearly doubled, and virtually all voice users are on GPT-Live. OpenAI has not published the figure the doubling started from, or the exact share of users still on Standard or Advanced. In July, the company said more than 150 million people a week talked to ChatGPT using voice or dictation, a figure that combines the two.
The evidence that GPT-Live improves on Advanced also comes from OpenAI’s own measurements. In July, the company published evaluations by human raters in which Live was “strongly preferred” over Advanced. It also published results from its own tests of reasoning and web search. OpenAI cites no independent test.
OpenAI promises to improve the depth and detail of answers, continuity across conversations and personalization. It has given no date.
What it means
Starting December 11, ChatGPT will have a single voice mode for most uses. Video and screen sharing stay on Advanced for now.
The change mainly affects people who stuck with Standard or Advanced. They lose the turn-by-turn exchange, in which ChatGPT waits for them to finish speaking. For Standard users, recordings will no longer be deleted right after transcription. People who use custom GPTs lose voice in them on December 11, when the GPTs themselves are retired.
OpenAI has not published enough data to check the two claims behind its decision. Some complaints, such as Live forgetting context, relate to an area the company says it wants to improve, though it has set no deadline. The complaint reported most often, that Live cuts in too quickly, is not among the improvements OpenAI has promised.
OpenAI has published no evaluation of Live in French, even though it added two French voices in October.
In 2025, OpenAI put the removal of Standard on hold a month after announcing it. The next ChatGPT release notes will show whether the December 11 date holds this time, and when GPT-Live will support video and screen sharing.
Sources
- Retiring Advanced and Standard Voice in ChatGPT, OpenAI Help Center, November 10, 2026
- ChatGPT release notes, OpenAI Help Center, entries from August 7, 2025 to November 10, 2026
- ChatGPT Voice, OpenAI Help Center, accessed November 16, 2026
- Introducing GPT-Live, OpenAI, July 8, 2026
- ChatGPT Business release notes, OpenAI Help Center, entries of July 23 and September 9, 2026
- ChatGPT Is Killing Off Standard Voice Modes and Users Are Frustrated, Android Headlines, November 11, 2026
- OpenAI is sunsetting ChatGPT Advanced and Standard voice next month, Android Authority, November 11, 2026
- OpenAI supprime deux modes vocaux de ChatGPT : certains utilisateurs le vivent très mal, LeBigData.fr, November 11, 2026
Header image: MACCUS chart, data from the ChatGPT release notes and the OpenAI Help Center, accessed November 16, 2026



