The ChatGPT Advanced Voice retirement is set for December 11, 2026. OpenAI’s GPT-Live help article says Advanced and Standard voice both retire that day, and anyone who has either one selected is switched to Live automatically. If your learner instructions say “tap the voice icon and choose Advanced,” they need rewriting before then.
This post covers what OpenAI’s help center says as of November 18, 2026, what it does not say, and a dated checklist for training teams that use ChatGPT voice for roleplay, language practice or learner demos. The checklist is our analysis, not OpenAI guidance. Everything about the product comes from OpenAI’s published documentation; this site has not tested Live.
When is the ChatGPT Advanced Voice retirement, and what replaces it?
OpenAI’s help center says Advanced and Standard voice both retire on December 11, 2026. If either is selected, ChatGPT switches the account to Live on that date. Live is the current free-form voice mode. The page gives no time of day or time zone for the cutoff, so treat December 10 as your last safe day.
One third-party write-up, published November 12, reports that OpenAI notified users on November 10. We could not open OpenAI’s separate retirement notice, so treat that date as reported, not confirmed.
| Voice mode | What OpenAI’s FAQ says it is | Status |
|---|---|---|
| Live | Natural, free-form voice powered by GPT-Live-1.1 or GPT-Live-1.1 Quip, depending on plan | Stays; the default after December 11 |
| Advanced | The previous real-time voice experience, used when you need mobile video or screen sharing | Retires December 11, 2026 |
| Standard | Turn-by-turn: transcribes your speech before generating a response | Retires December 11, 2026 |
What is Live voice, and how does it differ from Advanced Voice?
Live is built for natural back-and-forth speech. OpenAI says it listens and speaks at the same time, so a learner can interrupt, and it can use web search, memory, text, images and plugins in the same chat. It is designed mainly for one-on-one conversation, not several speakers.
Live is not brand new. OpenAI’s release notes show GPT-Live-1 arriving on July 8, 2026, and GPT-Live-1.1 with new voices on October 15, 2026. Some of your facilitators may already be using it.
The FAQ lists several behaviors that matter for training use:
- Preset ChatGPT personalities do not currently apply to Live.
- Live does not support video or screen sharing.
- You can have one voice conversation at a time.
- Changing the voice mid-conversation starts a new call in the same chat.
- If Voice detects the wrong language, set one under Settings, Voice, Language, or ask ChatGPT to switch.
Who can use Live, and what are the voice limits?
Availability depends on plan, workspace settings, region, app version and parental controls, according to the FAQ. Free accounts get limited access to GPT-Live-1.1 Quip, and most paid plans have rolling 24-hour caps. Voice works on web, iOS and Android, with a separate desktop option. Limits can change, so check before a cohort session.
| Plan | Voice in Chat limit (rolling 24 hours) |
|---|---|
| Free | Limited access to GPT-Live-1.1 Quip |
| Go and Plus | 3 hours |
| Pro ($100/month) | 15 hours |
| Pro ($200/month) | Unlimited, subject to abuse guardrails |
| Business Standard | 3 hours; extra use 1.25 credits per minute |
| Business Premium | 15 hours; extra use 1.25 credits per minute |
| Enterprise, Edu (credit-based) | 1.25 credits per minute |
The FAQ also lists K-12 Teachers at 3 hours, usage-based Enterprise at $0.05 per minute, and legacy Enterprise and Edu at 3 hours. Because limits are set per plan, a facilitator demoing on a Pro account and a learner practicing on a free account will hit different ceilings. For plan prices, see our ChatGPT pricing guide.
What is still unconfirmed about the ChatGPT Advanced Voice retirement?
Three things matter most to training teams and are not settled on the OpenAI page we could read: whether video and screen sharing survive after December 11, what happens to old Advanced and Standard conversations, and whether custom instructions carry over. As of November 18, 2026, none of the three is confirmed.
- Video and screen sharing. The FAQ says Live does not support them and Advanced does, on iOS and Android. The July 8 release note said subscribers could keep using them with Advanced. The third-party write-up says OpenAI’s retirement notice keeps them on the older mode until Live gains them. The FAQ also says Advanced retires December 11. We cannot reconcile these, so do not plan demos around either outcome.
- Existing conversations. The write-up says only that you can request a ChatGPT data export to keep history. Whether old voice chats stay visible is not confirmed.
- Custom instructions. The FAQ does not mention them. It says only that preset personalities do not currently apply to Live.
- Cutoff time. No time zone is stated.
Turn Every Unknown Into a Test Case
Each open item above is a yes-or-no you can answer by trying it on a real account. Record the answer, the date and the device in a change log at the bottom of each job aid, so a learner who sees different behavior can tell whether the document or the product changed.
How do you prepare for the ChatGPT Advanced Voice retirement?
Work in four moves: inventory every place voice is used, test each use on Live, rewrite the materials that change, and set a fallback for anything that fails. This is our analysis, not OpenAI guidance. From November 18, you have 23 days, and the dates below leave a few days of margin.
Step 1: Inventory. Search your LMS, job aids, facilitator guides, demo scripts and screenshots for “ChatGPT,” “Advanced Voice,” “Standard Voice” and “voice mode.” Log each use: learner practice, facilitator demo or self-study.
Step 2: Test on Live. Run each use on Live, on the devices your learners use, and mark pass, partial or fail.
Step 3: Update. Rewrite instructions that name the old modes, and retake screenshots that show the old picker.
Step 4: Fallback. For every fail, name a replacement activity and tell learners before December 11.
| Date | Task | Suggested owner |
|---|---|---|
| Fri, Nov 20 | Inventory finished and every item assigned | Instructional designer |
| Wed, Nov 25 | Every use tested on Live; pass, partial or fail logged | Designer and facilitators |
| Fri, Nov 27 | Keep, rewrite or replace decided per item | L&D lead |
| Fri, Dec 4 | Updated job aids, screenshots and instructions live in the LMS | Designer and LMS admin |
| Tue, Dec 8 | Fallbacks ready; learner notice sent | Program manager |
| Thu, Dec 10 | Final recheck on Live (cutoff time zone unconfirmed) | Facilitators |
| Fri, Dec 11 | Retirement day; watch for learner support questions | L&D operations |
| Mon, Dec 14 | Recheck unconfirmed items against OpenAI’s pages; close the change log | Instructional designer |
What should you test on Live for roleplay and language practice?
Test the behaviors OpenAI says differ: personas, interruption, language, voice changes and limits. A roleplay script that relied on a preset personality, or on a long uninterrupted learner turn, is the likeliest to behave differently. Run each script on web and on mobile before you sign it off.
- Persona. Does your setup prompt for the difficult customer or reluctant employee hold up in speech? Preset personalities do not apply, so the persona must live in your own instructions.
- Interruption. Learners can cut in. Check that the scenario still lets them finish a full answer, and that interruptions do not break scoring or debrief steps.
- Language. Set the language under Settings, Voice, Language and test every language you teach. OpenAI lists added voices for Brazilian Portuguese, French, German, Hindi, Japanese, Korean, Mexican Spanish and English (Bubbie). Spoken output is not reviewed localization; for regulated wording, see our guide to AI translation for training content.
- Sessions. One voice conversation at a time, and a voice change starts a new call. Brief learners so they do not restart mid-scenario.
- Privacy. OpenAI says audio and video clips are stored with the transcript for 30 days and are not used for training unless you choose to share them. Check that against your learner-data policy.
What fallback should you set if Live does not fit?
Pick the fallback by learning goal, not by tool. For spoken practice, a dedicated roleplay platform is one option; for screen-share demos, a recorded walkthrough avoids relying on a feature whose status is unconfirmed. Whatever you choose, test it the same way you tested Live.
Our comparison of AI roleplay tools for corporate training covers dedicated platforms. Voice reliability is a separate question; our post on AI voice agents training looks at what one published success-rate result does and does not show.
Write Fallbacks by Skill, Not by Tool
Word the learner instruction around the skill, such as “practice the objection-handling call aloud with an AI partner,” and put the current tool in a separate line you can swap. When a vendor changes modes again, you edit one line, not every course page.
What should you do next?
Start the inventory today. Search your LMS and shared drives for “ChatGPT,” “Advanced Voice” and “voice mode,” and assign each hit an owner and a test date from the table above.
Before you message learners, reread OpenAI’s GPT-Live FAQ. Its pages can change after November 18, 2026, and the video, history and custom-instruction questions may be answered by then.
FAQ
Q1. When is ChatGPT Advanced Voice retiring?
OpenAI’s help center says Advanced and Standard voice both retire on December 11, 2026. Anyone with either one selected is switched to Live automatically on that date. The page we read gives no time of day or time zone, so finish any voice-based training changes before December 11.
Q2. What is ChatGPT Live voice?
Live is ChatGPT’s free-form voice mode, powered by GPT-Live-1.1 or GPT-Live-1.1 Quip depending on plan. OpenAI says it listens and speaks at the same time so you can interrupt, and it can use web search, memory, images and plugins. It is designed mainly for one-on-one conversation.
Q3. Will video and screen sharing stop working on December 11?
Not confirmed as of November 18, 2026. OpenAI’s FAQ says Live does not support video or screen sharing and that Advanced does. The same FAQ says Advanced retires December 11. A third-party report says those features stay on the older mode until Live supports them. Check OpenAI’s help pages before planning demos.
Q4. Do I need a paid plan to use Live?
No, but free accounts get only limited access to GPT-Live-1.1 Quip. According to OpenAI’s FAQ, Plus, Go, Pro and K-12 Teacher plans include access with daily limits, and Business, Enterprise, Edu and Healthcare workspaces are eligible subject to workspace settings. Availability also varies by region and app version.
Q5. Will my old Advanced Voice conversations be deleted?
Not confirmed. The OpenAI FAQ we read does not say what happens to past Advanced or Standard voice conversations. One third-party write-up mentions only that you can request a ChatGPT data export to keep your history. If transcripts matter for training records, export them before December 11.
Q6. Does Live support custom personalities?
OpenAI’s FAQ says preset ChatGPT personalities do not currently apply to Live, though you can ask ChatGPT to change tone, pace or style during a conversation. The page does not mention custom instructions. If your roleplay depends on a persona, test it on Live and keep the setup prompt in your own materials.
Q7. How long does OpenAI keep voice recordings?
OpenAI’s FAQ says audio and video clips are stored with the transcript for 30 days. It also says clips are not used for model training unless you choose to share them. If learners practice with real scenarios, check that retention period against your organization’s learner-data policy.