ChatGPT will retire Advanced and Standard voice on December 11

ChatGPT has had three ways to talk out loud. Standard voice is the oldest. Advanced voice came next. GPT-Live, introduced in July 2026, is the newest and the default for most people.

On December 11, 2026, OpenAI will take Advanced and Standard voice out of the voice picker under Settings. If you have either one selected, ChatGPT will switch you to Live voice on its own. Both older options stay available until then.

OpenAI announced the change on November 10 from its official newsroom account. There was no long blog post. The post was short and said the company wants to simplify voice.

The old modes worked in a different way, and that explains why they feel different.

• Standard voice: three models in a row. One turned your speech into text. A second wrote the reply. A third read the reply out loud.
• Advanced voice: one model that handled audio directly. It still waited for you to finish speaking before it answered.
• GPT-Live: a model that listens and talks at the same time.

That last one is called full-duplex, which is a technical way of saying both sides can talk at once, like a phone call and unlike a walkie-talkie.

In real life With Advanced voice, you ask a question, wait, and then hear the answer. With GPT-Live, you can cut in with "wait, I meant Tuesday" and it adjusts without starting over.

OpenAI says GPT-Live decides many times a second whether to speak, keep listening, pause, interrupt or use a tool. It can say short phrases like "mhmm" or "got it" to show it is following along. You can interrupt it, take a moment to think or ask it to slow down.

When a question needs a web search or hard thinking, GPT-Live hands the work to a bigger text model in the background. At launch that model was GPT-5.5. The voice keeps the conversation going while that model works.

You can also pick a reasoning level, from Instant for quick answers to Medium and High for harder tasks. And the voice can show visual cards for things like weather, stock prices or sports scores. OpenAI also redesigned the nine voices.

By the numbers: in OpenAI's own tests at launch, people preferred GPT-Live-1 over Advanced voice 75.7 percent of the time. For the smaller GPT-Live-1 mini the figure was 69.2 percent.

OpenAI gives two reasons. First, voice use has nearly doubled since GPT-Live arrived in July, and almost everyone who uses voice in ChatGPT already uses the newer model.

Second, running three systems at once means three sets of voices, limits and behavior to build and maintain. Dropping two lets OpenAI put its effort into one.

OpenAI also says feedback from people who preferred the older modes will shape the next work on GPT-Live. It named response depth and detail, continuity across conversations and personalization.

Why it matters: this is a company choosing one voice path and putting its full effort behind it. If you use ChatGPT by voice, the one you will hear after December 11 is the one that gets all the improvements.

Two things keep Advanced voice alive for a while longer:

• Live video and screen sharing: these still run on Advanced voice. GPT-Live did not support them at launch, and OpenAI says they will stay on the older mode until the new one can handle them. No date has been given.
• Custom GPTs: voice chats with custom GPTs use Advanced voice until December 11. That is also the day custom GPTs reach their own scheduled retirement, as OpenAI helps creators move them to plugins. FedRAMP and some Enterprise workspaces have a later date.

Your old chats are safe. Conversations started in Advanced or Standard voice stay in your history. Reopen one and continue by typing or speaking. If you speak, GPT-Live answers.

When GPT-Live launched in July, it came in two sizes. GPT-Live-1 is for people on the Go, Plus and Pro plans. GPT-Live-1 mini is for free accounts. Both work on iOS, Android and ChatGPT.com.

OpenAI also said it would add API access soon, so developers could build voice features into their own apps. An API is a doorway that lets one piece of software talk to another.

Live voice models have had a known weakness. They answered from their own knowledge only. That left them far behind the best text models, so a voice chat could sound smooth and still give a weaker answer than typing the same question.

GPT-Live was built to close that gap. It splits the job in two: one part manages the conversation, and the other part does the thinking. OpenAI says the design keeps GPT-Live connected to the latest top models, so the voice can improve when the model behind it does.

OpenAI is not alone in this idea. Earlier this year, Nvidia released an open-source model called PersonaPlex that also listens and speaks at once. Open-source means anyone can download it and run it.

You have time, and a few small steps will make the switch easier:

• Try Live voice now: open Settings, find Voice and pick Live. Use it for a week on the things you usually ask by voice.
• Check your custom GPTs: if you talk to any of them by voice, plan for the December 11 date.
• Keep what you need: old chats stay, so there is no need to copy anything out.

Not everyone is cheering. On OpenAI's community forum, paying users have asked the company to keep Standard voice. Some say it is the mode that works best for them, and some say accessibility is the reason, including for neurodivergent users.

These are users' own words, and OpenAI has not said it will change course. What it has said is that it will use what it hears from fans of the older modes to improve GPT-Live.

It is a familiar tension. A newer model is better on average, while a small group of people had found the older one fits how they work.

• If you use Standard or Advanced voice: nothing breaks today. On December 11, you will be moved to Live voice, so try it before then.
• If you use video or screen sharing: these keep working on Advanced voice for now.
• If you rely on a custom GPT by voice: the Advanced voice option ends on December 11 along with those custom GPTs.
• If you miss the old voice: tell OpenAI. It says it is listening to that feedback.