Grok Voice Explained: App Voice Mode, Voices, API and Pricing
Grok Voice is xAI’s speech stack: the Grok app’s voice mode, the Grok Voice 3.0 speech-to-speech model, 28 voices, and APIs for text to speech, transcription and voice cloning. What each part does and costs.
Last updated:
"Grok Voice" means a few different things depending on who is saying it. This guide sorts them out: the voice mode in the Grok app, the Grok Voice 3.0 model behind it, the 28 voices, and the Grok Voice API that developers use for text to speech, voice agents, transcription and voice cloning.
1. Voice mode in the Grok app
The consumer side. In the Grok app (iOS, Android, grok.com), the microphone or waveform button next to the message box opens a live, spoken conversation. You can interrupt Grok mid-sentence, it responds with tone rather than a flat read, and it can look things up while talking.
Voice mode runs on the Grok Voice 3.0 speech-to-speech model since October 2026. The voice you hear is one of the original Grok voices: Ara, Eve, Leo, Rex or Sal. The app also offers personality presets that change how Grok behaves, separate from which voice speaks.
2. The Grok Voice 3.0 model
The engine. Grok Voice 3.0 is a speech-to-speech model: it takes audio in and produces audio out, with reasoning and tool use in between, instead of chaining transcription, a text model and a text to speech voice. xAI reports it at #1 on the Artificial Analysis Speech-to-Speech Index (86.9) and uses it for Tesla and Starlink support calls. Details and benchmarks are on the Grok Voice 3.0 page.
3. The 28 Grok voices
The sound. The Voice API ships 28 built-in voices; all of them speak all 20 supported languages, and all of them accept speech tags like [pause], [laugh] and <whisper>:
- Original five (also in the app): Ara, Eve, Leo, Rex, Sal.
- Added with the API: Altair, Atlas, Aurora, Carina, Castor, Celeste, Cosmo, Helios, Helix, Iris, Kepler, Liora, Lumen, Luna, Lux, Naksh, Orion, Perseus, Rigel, Sirius, Ursa, Zagan, Zenith.
Hear any of them with your own text on the voices page or straight from the homepage tool.
4. The Grok Voice API
The developer side, at x.ai/api/voice. Four capabilities:
| Capability | What it does | Price |
|---|---|---|
| Speech to speech | Real-time voice agents over WebSocket, with tools, search and SIP phone calls | $0.08 / min |
| Text to speech | POST /v1/tts: text in, MP3/WAV/PCM out, 28 voices, speech tags, timestamps | $15 / 1M characters |
| Speech to text | Transcription of files or live streams, with speaker diarization and entity recognition | $0.10 / hr REST, $0.20 / hr streaming |
| Custom voices | Clone a voice from about two minutes of audio, then use it in TTS and agents | Create in console; usage billed as TTS |
Enterprise notes from xAI's page: SOC 2 Type II, HIPAA-eligible with a BAA, GDPR data-processing agreements, zero-data-retention option, custom rate limits and SSO.
Guides on this site: Grok Voice API: text to speech endpoint, building a Grok voice agent, Grok voice clone, and the pricing calculator.
Which one do you need?
- Want to talk to Grok? The app. Nothing to set up.
- Want Grok's voices to read your text? Text to speech. Try it free here; in production it is $15 per million characters.
- Want a bot that answers calls or talks to users in your app? Speech to speech with Grok Voice 3.0, $0.08 per minute.
- Want your own voice (or a brand voice) in any of the above? Custom voices.
About this site
Grok Text to Speech is an independent tool built on the public xAI API so people can try the voices without an account. It is not affiliated with xAI. Facts on this page were checked against xAI's documentation and launch post on 16 October 2026; the official pages win if anything here is out of date.