ChatGPT's New Voice (GPT-Live) Explained: What's New and Do You Get It

The short version: ChatGPT's voice conversations got a major upgrade on July 8, 2026, and free users are included. OpenAI launched GPT-Live, a new voice model that powers a rebuilt ChatGPT Voice, and the headline change is simple to feel: it can listen and talk at the same time. Older voice modes waited for you to finish speaking before they replied, which made conversations feel stiff. GPT-Live flows -- it can jump in, murmur "mhmm" to show it is following, or stay quiet while you gather your thoughts. Who gets which version has since changed: on September 9, 2026 OpenAI reset the daily hour limits on every paid plan and pushed ChatGPT Go down to the smaller model. This page explains what GPT-Live is, what you now get on your plan, and how it differs from ChatGPT's older Advanced Voice Mode.

Key Takeaways

• GPT-Live launched July 8, 2026. It is a new voice model that replaces the engine behind ChatGPT Voice, built so it can listen and speak simultaneously instead of taking rigid turns.
• Free users are included. OpenAI is rolling it out "across consumer plans, including Free," on chatgpt.com and the ChatGPT iOS and Android apps in supported regions.
• Two versions, and the line between them moved on September 9, 2026. GPT-Live-1 mini is what Free and now ChatGPT Go get; the full GPT-Live-1 starts at Plus. OpenAI's release note says Go gets mini "replacing GPT-Live-1 access", so Go lost something it had.
• Voice is timed. Plus and Go get 3 hours a day, Pro at $100 gets 15 hours, Pro at $200 is unlimited, and Free gets "limited access" with no number. Before September 9 the limits were split by intelligence level on OpenAI's Voice help page.
• It feels like a real conversation. The full-duplex design means it can interrupt, acknowledge, or wait -- and it keeps text on screen, so you see a written transcript alongside the spoken reply.
• It can still search, remember, and show things. GPT-Live uses web search and memory, shows visual widgets, and handles text and images in the same conversation.
• No video or screen sharing yet. Those stay on Advanced Voice Mode for eligible subscribers for now. GPT-Live was not in ChatGPT Business, Enterprise or Edu at launch, but OpenAI's Voice help page now lists it for eligible workspaces.
• It reached the desktop app on July 23, 2026. ChatGPT Voice now works inside the desktop app's Work and Codex views, so you can start and steer a task by talking instead of typing.
• Since August 7, 2026 you can upload files during a voice conversation and use Voice inside Projects -- so you can hand ChatGPT a document and then just talk about it.
• Since August 31, 2026 a voice conversation can follow you out of the app on an iPhone, showing on the Lock Screen and in the Dynamic Island once you turn on Background conversations.

What Is GPT-Live?

GPT-Live is a new generation of voice model from OpenAI, and it now powers the ChatGPT Voice you tap into from the message box. The important change is architectural: OpenAI built GPT-Live on what it calls a full-duplex design, meaning it can listen and speak at the same time. In everyday terms, it behaves more like a person on a phone call than a walkie-talkie. It can react while you are still talking, drop in a quick "yeah" or "mhmm" to show it is paying attention, engage in fast back-and-forth, or simply stay quiet when you need a moment to think.

That is a real departure from how ChatGPT Voice worked before. The original Voice chained three separate models together -- one to transcribe your speech, one to think, one to speak -- which was slow and stilted. The newer Advanced Voice Mode was smoother but still turn-based: it waited for silence before replying, so a pause or a bit of background noise could make it cut in at the wrong moment. GPT-Live is OpenAI's fix for that rigidity.

What Can It Actually Do?

GPT-Live is more than a smoother talker -- OpenAI calls it its smartest voice model yet, and it plugs into the rest of ChatGPT. Spoken answers appear alongside streamed text on screen, so you get a readable transcript as it speaks. It can use web search and memory, show visual results through supported widgets (think maps or weather cards), and work with text and images in the same conversation. So you can ask it something out loud, watch the answer appear as text, and have it pull a live search result into the chat without breaking the flow.

For anything that needs deeper thinking or a web lookup, GPT-Live hands the hard part to OpenAI's latest frontier model behind the scenes and brings the result back when it is ready -- and it can keep chatting with you while that work happens. At launch, GPT-Live used GPT-5.5 as that background brain. OpenAI said it would update which model does that job as newer ones ship, and on September 9, 2026 it did: its release notes say "ChatGPT Voice can now use GPT-5.6 or GPT-6 Astra when it needs to search or reason through harder questions," and that you choose the model and reasoning effort "using the same controls as text chat." If "model" and "reasoning" are fuzzy terms, our plain-terms guide to what an AI model is doing may help.

Do You Get It, and on Which Plan?

Yes, on every consumer plan -- but which version you get, and for how long a day, changed on September 9, 2026. Every plan still has voice. What OpenAI did was replace its old limits, which were split by intelligence level, with one hours-per-day figure per plan, and move ChatGPT Go from the full model down to the smaller one. Here is the breakdown from OpenAI's release notes and OpenAI's Voice help page, re-read row by row on October 3, 2026 against that help page's newest archived copy, dated September 25, 2026. Every figure below still matched. The help page measures every limit over a rolling 24-hour period:

The Go row is the one that stings. OpenAI's wording is "Go: Up to 3 hours with GPT-Live-1 mini, replacing GPT-Live-1 access" -- so a Go subscriber who could hold a full GPT-Live-1 conversation in August now gets the smaller model for the same $8. Nothing about it was announced outside the changelog. If you pay for Go mainly to talk to ChatGPT, whether ChatGPT Plus is worth it is now a live question for you rather than an abstract one.

There is a genuine gain buried in the same entry, and it goes to Plus and Pro. Before September 9, once Plus or Pro used up its GPT-Live-1 hours, voice carried on with GPT-Live-1 mini for the hours that remained. OpenAI says that stops: "Plus and Pro no longer switch to GPT-Live mini after reaching a Voice limit." You get the full model up to your ceiling, and then you stop, which is easier to plan around than a quality drop you cannot see.

One control disappeared too. The separate Instant / Medium / High voice intelligence levels are deprecated. You now pick a model and a reasoning effort with the same controls you use for text chat, rather than a voice-only setting.

Two caveats worth knowing. First, GPT-Live does not support video or screen sharing, and OpenAI states that twice on its Voice help page in those words -- if you rely on pointing your camera at something or sharing your screen during a voice chat, ChatGPT keeps Advanced Voice Mode available for eligible subscribers to do that. This is the one place a rival is plainly ahead on the free tier: Gemini Live shares your camera and your full screen at no cost, on Android and iPhone alike. Second, this is a gradual rollout, so if you do not see it yet, it is still reaching your account. Make sure your app is updated and check again over the coming days.

One forward-looking line from launch day has not landed. OpenAI said on July 8, 2026 that it planned to bring GPT-Live to its API for developers soon. As of October 3, 2026 we can find no OpenAI surface announcing it, and the Codex pricing doc says the opposite for one surface at least: "ChatGPT Voice in Desktop isn't available via API." Nothing about this affects the app you use; it is only flagged here because "soon" is now 87 days old.

Talking To ChatGPT on Your Computer (July 23 Update)

Voice is no longer only for the phone in your hand. On July 23, 2026, OpenAI's release notes recorded that ChatGPT Voice became available in Work and Codex inside the ChatGPT desktop app on macOS and Windows. In plain terms: on the desktop app you can now start a task by speaking, interrupt it mid-flow, and ask Voice to kick off or coordinate work using whatever tools and permissions that view already has.

What that changes for an ordinary person is narrower than it sounds, so be clear on which thing you are using. Voice for normal conversation is the same feature it was on July 8, on the web and in the phone apps. This addition is about the desktop app's Work view -- the agent that goes off and produces documents, spreadsheets, and reports -- and Codex, which is for software work. Work is not on the Free or Go plans, so if you are on either of those, spoken conversation is what you get and this update does not reach you.

What changed (July 25, 2026): added this section after OpenAI's release notes confirmed the July 23 arrival of Voice in the desktop app's Work and Codex views. Nothing about GPT-Live on the web or in the phone apps changed.

Voice Can Now Read Your Files (August 7 Update)

The most useful thing to happen to ChatGPT Voice since launch is small and easy to miss. On August 7, 2026, OpenAI's release notes recorded that "GPT-Live in ChatGPT Voice now supports file uploads and Projects." In its own words: "You can now upload files in a voice conversation and analyze its contents or ask questions. You can also use voice in Projects, referencing recent project chats, sources, and project instructions."

In plain terms, a spoken conversation is no longer limited to what you can say out loud. You can hand ChatGPT a document mid-conversation and then talk about it -- read me the bit about cancellation, is this bill higher than last month's, what is this form asking me for -- instead of switching to typing to attach it. And if you keep related chats in a Project, Voice can now draw on that Project's files, past chats and instructions rather than starting cold.

Two honest limits on this one. OpenAI's entry names no plan gate and no country list, so this page does not claim one either way; and the file-upload limits that apply everywhere else in ChatGPT are not restated, so assume your plan's normal upload allowance still applies rather than a new one.

What changed (August 10, 2026): added this section after OpenAI's August 7 release-notes entry. Nothing else about GPT-Live changed -- the models, the plan split and the missing video and screen sharing are all as they were.

Voice Follows You Out of the App on iPhone (August 31 Update)

A voice conversation no longer stops mattering the moment you switch apps. On August 31, 2026, OpenAI's release notes recorded that "content from a Live voice conversation can now appear on your iPhone's Lock Screen and, on iPhones with Dynamic Island, in the Dynamic Island." It uses Apple's Live Activities, so the conversation shows up in the same place a food delivery or a timer does.

There is a switch to find first, and it is not on by default. OpenAI's instruction: "To continue a Voice conversation outside ChatGPT, turn on Background conversations in Settings > Voice." Once that is on, "Live Activities can show content from your voice conversation while you use another app or your phone is locked."

Two honest limits. OpenAI's entry names only the iPhone, and says nothing about Android either way, so this page does not claim Android has it or that it does not. And no plan is named, which on this feature's record means it is most likely on the same all-consumer-plans footing as the rest of Voice rather than a paid extra, but OpenAI has not said so.

What changed (September 5, 2026): added this section after reading OpenAI's August 31 release-notes entry, which we could not access when it was published. Nothing else about GPT-Live changed -- the models, the plan split and the missing video and screen sharing are all as they were.

Voice Can Use Your Connected Apps (September 23 Update)

You can now get things done by speaking, not just ask questions. On September 23, 2026, OpenAI's release notes recorded that "Live now supports plugins on web, iOS, and Android. You can use the plugins and connected apps available to your account during a Voice conversation and follow written responses in the chat."

The free plan is named, and that is the part worth knowing. OpenAI's exact sentence: "Free and Go users can use Voice in Chat with the plugins their plan supports." So this is not a paid upgrade. What you get depends on which plugins your plan already has, not on whether Voice itself is unlocked.

Two limits OpenAI states itself: "existing app connections, permissions, and usage limits apply," so speaking does not buy you more of anything, and Voice in Work is gated separately from Voice in Chat.

Is OpenAI Keeping Your Voice, and Does It Train On It?

The recordings are kept for 30 days and are not used for training unless you switch that on yourself. The transcripts are a different answer, and it is the one most people get backwards. OpenAI sets both out in the Data controls section of its Voice help page, read October 3, 2026:

So the useful summary is: the words are in scope by default, the recording of your voice is not. Both toggles live in Settings > Data Controls, and Improve the model for everyone has to be on before the audio ones appear.

Two limits on who can even make that choice. The clip-sharing option is for "Free, Plus, and Pro users in personal workspaces", and OpenAI says flatly that "users cannot share audio or video clips from Voice conversations in ChatGPT Business, Enterprise, or Edu workspaces." And deleting is one-way: "When you delete a chat, we also delete its associated audio and video clips within 30 days", but "archiving only removes the chat from your sidebar; it does not delete the chat or its associated audio or video clips." Archive is not delete.

GPT-Live vs Advanced vs Standard vs Dictation

There are now four things here, not three, and people mix them up constantly. Three are spoken conversations where ChatGPT talks back; dictation is the microphone button that just types what you say. OpenAI lists the first three together under Settings > Voice:

Which of the three you see "may depend on your plan, workspace settings, region, app version, and parental controls", in OpenAI's words, and on a linked teen account a parent can manage access to Voice entirely.

If your goal is a spoken conversation that feels natural, Live is the upgrade. If you specifically need to show ChatGPT your camera or screen during a voice chat, you will stay on Advanced for now. And if you just want to talk instead of type your messages, that is dictation -- a different button that got its own accuracy upgrade in June.

Three Smaller Limits Worth Knowing

All three are on OpenAI's Voice help page, none were announced, and each one catches people out:

• Live cannot reach your ChatGPT Library. "Live cannot currently find or add files from your ChatGPT Library." You can still attach a file to the chat by hand, which is what the August 7 change below enabled, but you cannot ask Voice to go and fetch one you saved earlier.
• Your chosen personality does not apply. "Preset ChatGPT personalities do not currently apply to Live." If you set ChatGPT to Friendly or Pragmatic, Voice ignores it. You can still ask it to change tone inside a conversation.
• It works in CarPlay. "ChatGPT is available in Apple CarPlay on supported iPhones", where you can start a voice conversation or continue a recent chat from the car screen. There is a Start automatically in CarPlay setting that appears once you have used it.

Should You Care?

Care if you talk to ChatGPT out loud, and especially if the old back-and-forth annoyed you. The single most common complaint about voice assistants -- that they cut you off, or freeze while you are mid-thought -- is exactly what the full-duplex design targets. For people who use voice hands-free while driving, cooking, or walking, a conversation that does not require rigid turn-taking is a genuine day-to-day improvement, not a spec-sheet one.

Try it if you gave up on ChatGPT Voice because it felt clunky -- this is the rebuild aimed at that. You do not need to do anything if you rarely use voice; nothing about typed chats changed. Stay on Advanced Voice Mode if you need video or screen sharing during a voice chat, since GPT-Live does not do those yet. Check what you are actually getting if you pay for Go, because since September 9 you are on the same voice model as the free tier.

The bottom line: GPT-Live made talking to ChatGPT genuinely better, and a version of it is still free. But September 9, 2026 redrew who gets what. Voice runs on a daily clock on every plan except Pro at $200, and ChatGPT Go quietly stopped being a paid voice tier in any meaningful sense -- it gets the same mini model as Free, capped at three hours. The full model starts at Plus. See how the assistants compare in our guides to the best free AI chatbots and whether ChatGPT Plus is worth it in 2026. Voice was part of a bigger week of ChatGPT changes -- see what else changed in ChatGPT's July app shake-up.

Corrections and Updates

September 19, 2026: this page said OpenAI published no daily voice limit before September 9 and that GPT-Live was not in Business, Enterprise or Edu; OpenAI's Voice help page, archived August 26, 2026, listed limits for every plan and GPT-Live for eligible workspaces.

September 11, 2026: this page said paying users get the full GPT-Live-1 and named ChatGPT Go as one of them. Go was moved down to GPT-Live-1 mini on September 9, 2026. OpenAI's release notes entry for that date reads "Go: Up to 3 hours with GPT-Live-1 mini, replacing GPT-Live-1 access." The same entry introduced daily voice limits on every paid plan, where this page had said none existed (superseded September 19: OpenAI's Voice help page already listed limits in August), and deprecated the Instant/Medium/High voice intelligence levels. The summary, the Key Takeaways, the free-tier FAQ answer, the availability table and the closing verdict were all corrected, and two FAQ answers were added for the new limits. The plan prices in the table were re-read from OpenAI's pricing docs on September 11, 2026. What GPT-Live is and how it differs from Advanced Voice Mode is unchanged.