How do you turn on voice mode in Claude?
You tap the sound wave icon. On the web and in Claude Desktop it sits in the lower right corner of a new chat. In the iOS and Android apps it sits beside the microphone icon in the text box. These steps follow Anthropic's voice mode help page and Engadget's walkthrough of 9 August 2026. We did not run voice mode ourselves for this page.

- Open claude.ai, Claude Desktop or the Claude app and start a new chat.
- Tap the sound wave icon, not the microphone. The microphone is dictation, which only types your words into the box.
- On a phone, pick a voice the first time. English has five voices (Buttery, Airy, Mellow, Glassy and Rounded) and three speeds (Slow, Normal, Fast), per Engadget.
- Speak. Claude answers when you stop talking, and the exchange is saved as text in your chat history.
- Claude stays in voice mode until you press Stop, which sits in the lower right corner of the chat window on web and desktop. To change the voice later, go to Settings > Capabilities > Voice settings on web and desktop, or tap the settings button at the bottom left during a call on mobile.
Claude has two ways to listen. Hands-free is the default: Claude listens all the time and replies at natural pauses, and if it cuts in on you, keep talking and it stops. Push-to-talk makes you hold a button while you speak and let go when you finish. Anthropic points to push-to-talk for noisy places and says hands-free works best in a quiet room. Voice and text share one conversation, so you can type a URL or a line of code mid-call and speak the rest.
Voice mode is built to work best from a phone, in Anthropic's words. It works in any conversation in the new Claude experience, but not inside Claude Cowork or Claude Code. How to use Claude covers the rest of the app, which this page leaves alone.
Which plans, models and languages does Claude voice mode have, and what are the limits?
Every plan has it, and every plan shares the same pot. Anthropic's page says "Voice conversations count toward your regular usage limits based on your subscription plan." There is no voice-minutes number to look up and no separate meter to run out of. A long call on Opus drains your allowance the way a long text chat on Opus would. If voice stops mid-call, Anthropic's own troubleshooting advice is to check your usage limits, then re-enter voice mode.
The chart is a compilation from three vendor pages, and it holds as of November 2026. OpenAI's voice page says "In Chat mode, Live usage is measured over a rolling 24-hour period," gives Free "limited access" to the Quip model, and makes the $200 Pro tier unlimited subject to abuse guardrails. Google's Gemini Live page states no usage limit at all. Anthropic doesn't either, and that's the point: Claude voice has no hour count to run out of, only the same limit your text chats draw on. ChatGPT voice mode limit has every OpenAI figure and the reset rule.
Models are where plans matter. Until 23 July 2026 voice ran only on Haiku. TechCrunch reported that day that you can now pick Opus, Sonnet or Haiku, and that voice starts on the model you last used in text chat. Anthropic's page adds that voice moves to the latest generation of that model, and that "Claude Fable is not currently available in voice mode." Anthropic's page lists no model per plan. TechCrunch reported free users limited to Haiku and one connected app, while Engadget later reported Haiku and Sonnet on Free and Opus only on a paid plan. Anthropic also lets you switch models with the model selector during a call. Check the model picker in voice mode to see what your account gets.
Languages are in beta outside English. Anthropic's page says voice supports "many more languages" without printing a list, so the counts come from the press and they disagree. TechCrunch counted 10 on 23 July 2026, eWeek counted 11 four days later, and Engadget counted 14 on 9 August 2026. Set yours at Settings > Capabilities > Voice > Spoken language, or Settings > General > Voice > Language. Changing the app's display language does not change the voice language, and Claude doesn't detect a switch on its own. Say "switch to German" out loud first.
What can Claude voice mode do, and what can't it do?
It can talk through a problem and reach the apps you have connected. Anthropic's page names web search plus Gmail, Google Calendar, Google Docs and Slack, and Free accounts get one connected tool while paid plans get more. TechCrunch also lists Canva and Notion. Connected tools work in voice the way they do in text chat and follow the rules of your plan. Four uses fit that shape, and the prompts below are examples written for this page, not tested runs.
- Rehearse a meeting.
I have a pitch to a client on Thursday. Act as the client. Ask me one hard question at a time and tell me after each answer what was vague. - Get through the inbox while walking.
Check my email from today. Read me the three that need a reply, then draft each reply aloud and wait for me to say yes before you save it. - Move a meeting.
Find my next free hour on Friday and tell me what is in the slot before it. Do not move anything yet. - Think out loud before writing.
I am going to ramble about the report. Do not interrupt. When I say done, give me a five-line outline.
What it can't do is just as concrete. Voice is turn-based: Claude listens, waits for you to stop, thinks and then speaks, so a pause mid-thought can be taken as the end of your prompt. Anthropic recommends asking multi-part questions one at a time. Some results can't be shown on screen in voice mode, and using several tools at once adds a short delay. Claude voices come from a limited preset list, which Anthropic says is meant to prevent cloning or impersonation. And voice isn't available in Cowork or Claude Code, where only dictation works.
Privacy is on the plain side. Anthropic says "By default, we don't keep the audio from your voice conversations": speech becomes text, the recording is deleted and the transcript stays in your chat history. OpenAI's help page says ChatGPT keeps audio clips from Live conversations for 30 days. On Free, Pro, Max and Ultra you can opt in to share recordings to improve voice mode and dictation. Enterprise owners who want voice off for the whole organization have to ask Support.
Is Claude voice mode better than ChatGPT voice or Gemini Live?
It depends on what you do with the call. Claude wins on one thing, there being no voice clock, and loses on how natural the conversation feels. ChatGPT Live handles speaking and listening at once, which OpenAI's help page confirms ("listen and speak at the same time"), while Engadget describes Claude as turn-based. Gemini Live is the one with a camera. The table is compiled from each vendor's help pages on 15 November 2026, for personal plans.
| Claude | ChatGPT | Gemini | |
|---|---|---|---|
| Where it runs | iOS, Android, Claude Desktop, web | Web, iOS, Android | Gemini phone apps (Google's steps are for Android), not the Gemini web app |
| Free plan | Yes, in beta | Limited access to the Quip model | The page does not say |
| Voice allowance | None separate, draws on your usage limit | Rolling 24 hours: Go 3 hours, Plus 3, Pro $100 15, Pro $200 unlimited | None stated |
| Camera and screen share | Not on Anthropic's voice page | Live has neither; the older Advanced voice did, and it retires on 11 December 2026 | Camera and full-screen sharing |
| Tools by voice | Gmail, Google Calendar, Google Docs, Slack, web search | Web search, memory, plugins and connected apps | Calendar, Keep, Tasks, Home, Flights, YouTube, Spotify |
| Models | Opus, Sonnet, Haiku | GPT-Live-1.1 and Quip | Not stated |
Pick Claude if you already pay for it and want voice for thinking and for Gmail or Slack work without watching a meter. Pick ChatGPT if you want the most natural back and forth and don't mind a 3-hour daily ceiling on Plus. Pick Gemini if you want to point a camera at something and ask about it, and you're on a phone. None of the three is a clear winner on speech quality: the sources that compared them called ChatGPT and Google more natural than Claude, and Anthropic made no change to the underlying voice model on 23 July, per TechCrunch.
Whizi, which publishes this page, has its own voice mode on the website, and it isn't Claude's. It runs on the Gemini Live API. The browser opens a connection straight to Google, so no audio passes through Whizi's backend. A session ends at 15 minutes, there is a single warm preset voice, saved memory is loaded when a session starts, and voice time is metered per plan, at 80 minutes a month on Pro. If you want Claude talking, use Claude's own app. The reason to look at Whizi is the typing side: Pro carries Claude, GPT and Gemini in one app, its first week is $1.99, and the free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5, no card. The voice mode docs cover how sessions are counted.
Why is Claude voice mode not working?
Most failures come down to the room, the microphone permission or the connection. Anthropic's troubleshooting list covers all three. Work down it in this order, because the first two take ten seconds each.
- Recognition is poor or Claude cuts you off: move somewhere quieter or switch to push-to-talk, and check that the app has microphone permission in your phone or browser settings.
- Audio is choppy or replies drop: check your connection, close heavy apps and make sure the battery isn't low.
- Claude stops answering: check your usage limits, since voice draws on them, then leave voice mode and start it again.
- The language is wrong: set the spoken language in Settings > Capabilities > Voice, because Claude won't detect it.
- The voice option is missing on an Enterprise account: an owner can disable it, so ask your admin.
If the sound wave icon is missing altogether, you're probably in Cowork or Claude Code, where voice mode doesn't exist. Or you're on an old app build, so update first. If everything above checks out and Claude itself seems down, Claude not working walks through the status page and the fixes in order.
Does Claude voice mode work on desktop and in Claude Code?
It works on the Claude Desktop app and on claude.ai in a browser, with the same sound wave icon. Claude Code is different: Anthropic's voice mode page says "In Claude Cowork and Claude Code, dictation is available but voice mode isn't." Dictation types your speech into the prompt box. It doesn't talk back.
Claude Code's dictation has its own commands, from Anthropic's Claude Code docs. Run /voice to turn it on and it checks your microphone. /voice hold sets hold mode, where you hold Space to record. /voice tap sets tap mode, where one tap starts and a second tap sends. /voice off switches it off. It needs a claude.ai sign-in, so it fails with an API key, Bedrock, Google Cloud or Foundry, and it doesn't work in cloud or SSH sessions. The docs list 20 dictation languages, and the setting follows your language value.
One surprise sits in those docs. They state that "Transcription does not consume Claude messages or tokens and does not count toward the limits shown in /usage." So the same company charges your limit for a spoken Claude conversation and charges nothing for speaking a prompt into Claude Code. In tap mode, recording stops by itself after 15 seconds of silence or two minutes in total. For the coding side, Claude Code commands has the rest of the slash list.
- Tap the sound wave icon, not the microphone beside it: the microphone only dictates into the text box
- Choose push-to-talk in noisy rooms and hands-free in quiet ones, and set the spoken language in Settings before you start
- Check the model picker in voice mode: Engadget reports Opus needs a paid plan, and Claude Fable is not available in voice
- Expect voice to spend your normal usage limit, because Anthropic publishes no separate voice allowance
- Connect only the apps a call needs, since Free accounts get one connected tool
- In Claude Code use
/voicefor dictation, which Anthropic says does not count toward/usage
Frequently asked questions
Is Claude voice mode free?
Yes. Anthropic lists voice mode as a beta feature on Free, Pro, Max, Ultra, Team and Enterprise plans, and voice conversations count toward the usage limit of whatever plan you have. Free accounts get one connected tool, and press coverage says they are limited to Haiku or to Haiku and Sonnet, with Opus on paid plans.
Does Claude voice mode have a time limit?
Anthropic publishes no voice time limit. It says voice conversations count toward your regular usage limits, so a long call ends when your allowance does, as a long text chat would. ChatGPT is different: OpenAI caps voice at 3 hours per rolling 24 hours on Plus and Go and 15 hours on Pro $100.
Can Claude voice mode use Opus?
Yes. Engadget reports Opus needs a paid plan. Anthropic opened Opus and Sonnet to voice on 23 July 2026, after voice ran only on Haiku, and voice starts on the model you last used in text chat. Claude Fable is the exception: Anthropic says it is not currently available in voice mode.
Which languages does Claude voice mode support?
Anthropic says voice mode supports many languages on every plan, with languages other than English in beta, and prints no list. Press counts differ: TechCrunch reported 10 on 23 July 2026 and Engadget reported 14 on 9 August 2026. Set yours under Settings > Capabilities > Voice > Spoken language.
Does Claude keep recordings of my voice chats?
Not by default. Anthropic says it doesn't keep the audio from voice conversations: your speech becomes text, the recording is deleted and the transcript is saved in your chat history. On Free, Pro, Max and Ultra you can choose to share recordings to help improve voice mode and dictation.