ChatGPT Live Dictation Explained: Text As You Talk, No 10-Minute Limit, Every Plan
OpenAI launched Live Dictation on November 18, 2026. Your words now appear as text while you are still speaking instead of after you stop, you can tap the waveform to hide the text, and the 10-minute recording cap is gone. It is rolling out globally to every ChatGPT plan, free included, on iOS, Android, web and desktop.
ChatGPT's voice typing now writes as you speak. On November 18, 2026 OpenAI announced Live Dictation, which turns your speech into text while you are still talking instead of transcribing it after you stop, lets you tap the waveform to hide the text, and removes the 10-minute recording limit. It is going to every plan, free included, on iOS, Android, web and the desktop app. One thing to know before you use the new, longer recordings: the audio is kept with the chat, and there is a switch for that.
What Is Live Dictation?
It is the same microphone button behaving differently: text appears while you speak rather than after. OpenAI's release notes, read live on November 18, 2026 with the page stamped "Updated: 2 hours ago", describe it in four sentences. Verbatim: "Dictation in ChatGPT is getting faster and more reliable. With Live Dictation, your speech turns into text as you talk, so you spend less time waiting for your message to finish transcribing. Tap or click on the waveform to show or hide text as you go. We've also removed the 10-minute recording limit." That is the whole of what OpenAI has published about it.
Four things change, and only one of them is a limit lifting:
| What changes | What it means when you use it |
|---|---|
| Speech becomes text as you talk | No pause at the end while it catches up. You can see mistakes while you are still speaking and correct them in the same breath |
| A waveform you can tap | Tap or click it to hide the text if watching it appear is distracting, then tap again to see it |
| The 10-minute recording cap is gone | You can dictate something long in one go. OpenAI has not published a replacement number |
| "Faster and more reliable" | OpenAI's words, with no figure attached. Treat it as a claim rather than a measurement |
Do You Get It, and Do You Have It Yet?
Every plan gets it, and "rolling out" means you might not have it today. OpenAI's wording is "rolling out globally to all ChatGPT plans on iOS, Android, web, and the desktop app." That is the broadest availability sentence OpenAI writes: no country list, no plan carve-out, no device exclusion.
What no Google-style help page will tell you is whether it has reached you, and that is worth being straight about. Both of OpenAI's dictation help pages were read live on November 18, 2026 and neither mentions Live Dictation: the Voice Dictation FAQ is 1,641 characters and stamped "Updated: last month", and ChatGPT Voice is 15,179 characters and stamped "Updated: 6 days ago". So the release note is currently the only vendor document that names the feature. The test is in the app: tap the microphone and speak. If words appear while you are still talking, you have it. If they appear after you stop, you do not, yet.
The Part Worth Reading Twice: Longer Recordings Are Kept
Removing a recording cap also removes a cap on how much of your voice OpenAI stores, and the switch for it is not in the dictation screen. This is not a new policy and OpenAI has not changed it; it just matters more now that a recording can run past ten minutes. From the Voice Dictation FAQ, read live November 18, 2026, verbatim:
- The audio is stored, not just the text. "Audio from dictation will be retained for as long as the chat is part of your chat history."
- Deleting the chat deletes the clip, with a delay and exceptions. "When you delete the chat, we'll also delete the associated audio clip within 30 days unless we need to keep it for security or legal reasons."
- It may train the models, if you opted in. "If you are a consumer user and have chosen to share audio to improve our models, we may train on audio from dictation." Business users are excluded by default.
- The off switch is in Data Controls. "You can stop sharing through the data controls page in your ChatGPT settings. Just disable the 'Include your audio recordings' toggle, or turn off 'Improve the model for everyone' entirely." OpenAI adds that this "will stop sharing all audio across dictation and voice chat", so it is one switch for both.
What to do about it: if you are about to start dictating long passages, open Settings then Data Controls first and look at "Include your audio recordings". Turning it off stops the training use. It does not stop the clip being stored with the chat; deleting the chat is what does that.
Live Dictation or Voice? OpenAI Now Answers This Itself
Use dictation to write a message, use Voice to have a conversation. OpenAI's ChatGPT Voice page, read live November 18, 2026, puts it plainly: "If you want to turn a single recording into editable text instead, use ChatGPT Dictation", and later, "Use ChatGPT Dictation when you want to record a prompt, review and edit its transcription, and then send it as text." Dictation gives you something you can edit before sending. Voice talks back and its transcript, in OpenAI's words, is "not verbatim".
The History: Why Dictation Was Already Better Than You Remember
Live Dictation is the third dictation change in 2026, and the earlier two are why the accuracy complaint may already be out of date for you. On June 26, 2026 OpenAI swapped in a new speech-to-text model for dictation across every plan, including free, and on August 7, 2026 it added an automatic retry on iPhone when a dictation drops. The rest of this page covers those two, and they still hold: the accuracy gains are in the model Live Dictation now streams from. It transcribes more accurately across languages and accents, handles people who switch languages mid-sentence, and holds up better in noisy places. Nothing about how you use dictation changed: same button, same steps, better results. This page explains what moved, who gets it, and how it differs from ChatGPT's talk-back Voice Mode.
Key Takeaways
- Live Dictation shipped November 18, 2026, and it is free on every plan. OpenAI's words: "rolling out globally to all ChatGPT plans on iOS, Android, web, and the desktop app." No country list, no paid tier, nothing to switch on.
- Your words appear while you are still speaking. "With Live Dictation, your speech turns into text as you talk, so you spend less time waiting for your message to finish transcribing." You can fix a misheard word without stopping.
- The 10-minute recording limit is gone, and OpenAI published no replacement. So there is no stated cap as of November 18, 2026.
- You can hide the text while you talk. "Tap or click on the waveform to show or hide text as you go."
- "Rolling out" is OpenAI's word, and no help page says who has it. Both of OpenAI's dictation help pages were read live on November 18, 2026 and neither names Live Dictation. The release note is the only vendor document that does, so the only reliable check is tapping the microphone yourself.
- The longer recording is stored with your chat. Not new policy, newly relevant: dictation audio is kept "for as long as the chat is part of your chat history", deleted within 30 days of deleting the chat, and may be used for training if you are a consumer who opted in to sharing audio. Switch: Settings, then Data Controls, then "Include your audio recordings".
- "Faster and more reliable" comes with no number. OpenAI gave a figure for the June 2026 model swap and has given none for this. Read it as a claim.
- A new dictation model shipped June 26, 2026. OpenAI replaced the speech-to-text engine behind ChatGPT's microphone button. It is a behind-the-scenes swap, so the button and the steps are unchanged.
- Everyone gets it, free included. The new model rolled out "across all plans." OpenAI's own current plan enumeration, in the July 6, 2026 release notes, is "Free, Go, Plus, Pro, Business, Enterprise, ChatGPT Edu, and K-12" -- all of them, with nothing to enable.
- Accuracy is up. OpenAI says word error rate -- the standard measure of transcription mistakes -- was at least 10% lower for the top languages tested than the previous model.
- Better with accents and mixed languages. It improved on Japanese, Korean, Chinese, Urdu, Vietnamese, accented English, and long-form Spanish, French, Italian, and Portuguese, including when a speaker switches languages mid-sentence.
- Tougher in the real world. It holds up better in noisy public spaces and workplaces, with quiet or whispered speech, and when you dictate strings of letters and numbers.
- This is dictation, not Voice Mode. The update improved the type-by-voice feature, not the spoken conversation feature.
What Changed in ChatGPT Dictation?
OpenAI replaced the model that converts your speech into text, and left everything else alone. In the June 26, 2026 release notes, OpenAI describes rolling out "a new speech-to-text model for dictation in ChatGPT across all plans" and calls it "a behind-the-scenes update, so there are no changes to how you use dictation." In plain terms: the microphone button you already use now runs on a smarter engine. You tap it, speak, and the text that lands in the message box is more likely to match what you actually said.
One follow-up since, on iPhone. OpenAI's August 7, 2026 release notes add, under iOS: "Dictate with fewer interruptions. If dictation fails because of a connection issue or timeout, ChatGPT automatically retries once, helping you finish recording without starting over." That is the fix for the case where a patchy signal used to lose what you had just said. It is iOS only in OpenAI's wording; the Android note from August 14, 2026 is a separate convenience, keeping the dictation button available after you add text or attachments.
The gains OpenAI lists are practical rather than flashy. Transcription accuracy improved across languages and accents, including the awkward case where a multilingual speaker mixes or switches languages inside one sentence. OpenAI names specific gains in Japanese, Korean, Chinese, Urdu, Vietnamese, accented English, and long-form Spanish, French, Italian, and Portuguese. Dictation is also described as more robust in noisy public spaces and workplace environments, with quiet or whispered speech, and when transcribing combinations of letters and numbers -- the exact situations where older voice typing tends to produce gibberish.
How Much Better Is It, Really?
OpenAI gives one hard number: word error rate at least 10% lower for the top languages tested. Word error rate is the standard yardstick for transcription -- it counts how many words a system gets wrong (inserted, deleted, or swapped) against a correct transcript. A 10% reduction means roughly one in ten of the mistakes the old model made are gone. That is a meaningful bump for something you use hands-free, but note the framing carefully: it is a relative improvement over OpenAI's previous production model, measured on the languages OpenAI tested, not a promise of flawless transcription. If you dictate in a well-supported language in a quiet room, you may not notice much; if you dictate with an accent, in a second language, or on a busy street, this is where the difference shows up.
| Situation | What OpenAI says improved |
|---|---|
| Non-English languages | Gains in Japanese, Korean, Chinese, Urdu, Vietnamese |
| Accented and long-form speech | Accented English; long-form Spanish, French, Italian, Portuguese |
| Switching languages mid-sentence | Better for multilingual speakers who mix languages |
| Noisy places | More robust in public spaces and workplaces |
| Quiet or whispered speech | More robust |
| Letters and numbers | Better at transcribing combinations |
| Overall accuracy | Word error rate at least 10% lower on top languages tested |
Availability by Plan, for Both Changes
You get both, whatever you pay. The June 2026 model swap went out "across all plans" as a server-side change, and OpenAI's November 18, 2026 wording for Live Dictation is "all ChatGPT plans". The difference is that the model swap was finished when announced and Live Dictation is still described as rolling out.
| Plan | Gets Live Dictation and the current model? | What you do to enable it |
|---|---|---|
| ChatGPT Free | Yes | Nothing -- it is automatic |
| ChatGPT Go | Yes | Nothing -- it is automatic |
| ChatGPT Plus | Yes | Nothing -- it is automatic |
| ChatGPT Pro | Yes | Nothing -- it is automatic |
| ChatGPT Business | Yes | Nothing -- it is automatic |
| ChatGPT Enterprise | Yes | Nothing -- it is automatic |
| ChatGPT Edu and K-12 | Yes | Nothing -- it is automatic |
This is unusual and worth flagging: most model upgrades in 2026 arrive gated behind a paid tier or a rollout list. This one did not. If you have been avoiding voice typing because it garbled your words, it is worth trying again -- the thing that annoyed you is the exact thing OpenAI says it fixed.
Dictation vs Voice Mode: Which One Got Better?
Dictation and Voice Mode are two different features, and only dictation changed here. Dictation is the microphone button that turns your speech into text inside the message box, so you can read, edit, and send it like anything you typed. Voice Mode is the separate feature where you have a spoken conversation and ChatGPT talks back out loud. People conflate them constantly, so to be exact: this June 26 update improved the speech-to-text behind dictation. It did not touch how ChatGPT speaks to you, its voices, or how Voice Mode conversations work.
Why does the distinction matter for you? Because if your complaint is "ChatGPT mishears me when I try to type by voice," this update targets that. If your complaint is "ChatGPT's spoken replies sound robotic" or "Voice Mode cuts me off," that is a different feature -- and OpenAI addressed it separately in July with GPT-Live, a rebuilt ChatGPT Voice that can listen and talk at the same time.
Should You Care?
Care if you type by voice, and especially if you do it in a second language or on the move. For everyday English dictation in a quiet room, the difference will be subtle -- the old model was already decent. The people who feel this most are multilingual speakers, anyone with an accent the old model stumbled on, and anyone who dictates in noisy real-world conditions. For them, "at least 10% fewer errors" is the difference between voice typing being usable and being more trouble than the keyboard.
Try it again if you gave up on ChatGPT dictation because it mangled your words -- this is aimed squarely at that. You do not need to do anything if you are happy with it already; you are on the new model regardless. Look elsewhere if what you actually want is a better spoken conversation, since that is Voice Mode, not dictation.
Note also what this update is not. OpenAI improved how accurately ChatGPT hears you; it still writes down what you said. Google's Rambler on the Pixel 11 takes the other route and rewrites you, stripping filler words and punctuating your sentences before the text appears. If you want your exact words, ChatGPT's approach is the right one. If you want a finished message, it is not.
One limit is worth naming, because a rival closed it in July. ChatGPT's dictation types into ChatGPT's own message box, so you still copy the result into whatever you were actually writing. Google's Speak to Window feature for the Gemini Mac app, switched on July 29, 2026, drops cleaned-up dictated text straight into any app on your Mac instead. If most of your typing happens in email, documents, or a browser rather than in a chatbot, that is the more useful shape.
The bottom line: ChatGPT voice typing is now live-updating, uncapped in length, and free on every plan, and OpenAI has documented all of that in a single release note and nowhere else. Use it if you dictate anything longer than a sentence, because watching the text appear is the difference between re-recording a paragraph and fixing one word. Check Settings then Data Controls before you start dictating long passages, since a recording with no time limit is also an audio clip with no length limit sitting in your chat history. Do not expect a measured accuracy gain from this one: the numbers OpenAI published belong to the June 2026 model change below, and "faster and more reliable" is the only claim attached to Live Dictation itself.
On the earlier change: OpenAI made a free, universal, no-effort upgrade to a feature millions of people use daily and barely announced it. If you decide between assistants partly on how well voice input works, this narrows ChatGPT's gap with phone-native dictation -- see how the assistants stack up in our guides to the best free AI chatbots and whether ChatGPT Plus is worth it in 2026.
Corrections and Updates
November 18, 2026: this page was about the June 2026 dictation model and said nothing about Live Dictation, which OpenAI shipped today and which changes what the microphone button does. Rebuilt around it rather than published as a second dictation page, per the roadmap rule preferring a ranking URL over a near-duplicate slug, which exp-001 measured. Sourced entirely from OpenAI's release notes, read live November 18, 2026 with the page stamped "Updated: 2 hours ago", whose November 18 entry is quoted in full on this page. Title, summary, SEO fields, four new FAQ answers, the lede, five Key Takeaways, four new sections and the verdict. sourcePublishedAt moved to 2026-11-18; publishedAt is unchanged at 2026-07-07, which is when this page first published.
Two things were deliberately not claimed. First, a new recording length: OpenAI removed the 10-minute cap and published no replacement, so this page says there is no stated cap rather than naming one. Second, that you have the feature: OpenAI says "rolling out", and both of its dictation help pages return zero matches for "Live Dictation" -- the Voice Dictation FAQ (1,641 characters, "Updated: last month") and ChatGPT Voice (15,179 characters, "Updated: 6 days ago"). A release note is a durable vendor source for what shipped; it is not a surface that says who has it, and this page says so instead of guessing.
The June 2026 material below was re-verified at the same time and nothing in it moved. ChatGPT Voice first returned a 403 with a 41-character body, which is a failed read rather than a finding; re-fetched after a pause it answered 200 at 15,179 characters.
October 25, 2026: the availability table on this page had no row for K-12. OpenAI enumerates its plans as "Free, Go, Plus, Pro, Business, Enterprise, ChatGPT Edu, and K-12," so Edu and K-12 now get their own row rather than being folded into Enterprise. Re-verified today: the June 26 entry is unchanged verbatim and no later release note has touched the dictation model.
September 3, 2026: nothing here, and Google now does a third thing with your voice that this page should name.
August 27, 2026: nothing here, and the rival's model finally has a name and a number.
August 22, 2026: nothing here, but a rival now solves a different problem.
August 15, 2026: nothing about the model, one more Android convenience.
August 13, 2026: the August 11 fix on this page corrected the availability table but missed the Key Takeaway, which still said "free, Plus, Pro, Team, and Enterprise" -- the same retired plan name, on the element AI answer engines quote most.
August 11, 2026: the availability table on this page listed "ChatGPT Team", a plan name OpenAI no longer uses, and left out ChatGPT Go entirely -- so a Go user could not find their own plan on a page whose whole point is that everyone gets this.
Read next
Keep up without the jargonGemini Speak to Window Explained: Talk to Your Mac in Any App
Google switched on Speak to Window in the Gemini app for macOS on July 29, 2026. Long-press the Fn key, talk, and clean text lands at your cursor in whatever app you are in. It is free and needs no subscription, but the app runs only on Apple Silicon MacBooks with macOS Sequoia 15 or later.
ChatGPT Is Back on WhatsApp in Europe: What It Means for You
On July 13, 2026, ChatGPT returned to WhatsApp across Europe (the EEA), after the EU ordered Meta to let rival AI assistants back on for free. You message it at the 1-800-CHATGPT contact with no account needed. OpenAI announced only the EEA and says availability follows your WhatsApp number's country code.
GPT-5.6 Sol Explained: What's New and What It Means for You
GPT-5.6 is here. On July 9, 2026, OpenAI released its flagship Sol plus the lower-cost Terra and Luna. Paid plans get Sol; Free and Go get Luna. From September 14, 2026 ChatGPT no longer switches from Instant to Thinking by itself on Plus and Pro, and the Higher intelligence setting is gone from the web on those plans.
Frequently Asked Questions
Is ChatGPT Live Dictation free?
Yes. OpenAI says Live Dictation is rolling out globally to all ChatGPT plans, which includes the free tier, on iOS, Android, web and the desktop app. There is no upgrade to buy and no setting to switch on.
What is Live Dictation in ChatGPT?
It is the microphone button, changed so your words appear as text while you are still speaking rather than after you stop. OpenAI says you can tap or click the waveform to show or hide the text as you go, and it has removed the old 10-minute recording limit.
How long can I record with ChatGPT dictation now?
OpenAI has removed the 10-minute recording limit and has not published a new number. So there is no stated cap as of November 18, 2026. Remember that a longer recording means a longer audio clip kept with that chat.
Do I have Live Dictation yet?
You may not. OpenAI calls it rolling out globally rather than live for everyone, and as of November 18, 2026 neither of its dictation help pages mentions Live Dictation at all, so there is no vendor page that tells you who has it. Tap the microphone and watch whether text appears as you speak.
What actually changed in ChatGPT dictation?
OpenAI replaced the model that turns your speech into text. It transcribes more accurately across languages and accents, handles multilingual speakers who switch languages mid-sentence, and copes better with noisy rooms, quiet or whispered speech, and strings of letters and numbers.
Is dictation the same as ChatGPT Voice Mode?
No. Dictation is the microphone button that turns your speech into text in the message box, so you can read and edit it before sending. Voice Mode is a spoken back-and-forth conversation where ChatGPT talks back. This update improved dictation, not Voice Mode.
Do I need to update the ChatGPT app to get it?
No. The change is server-side, so it applies the next time you use dictation regardless of app version. OpenAI describes it as a behind-the-scenes update with no changes to how you use dictation.