How to use Grok Voice 3.0 in Grok Bot

The documented path for Grok Voice 3.0 in Grok Bot: live voice chat, dictation, voice memos, settings, Team Bots, and the Voice Agent API.

The sections below follow the notes xAI has published on Grok Voice 3.0 and on voice in Grok Bot: the model, live voice chat, dictation, voice memos, desktop and mobile, settings, Team Bots, hanging up, and the Voice Agent API.

What Grok Voice 3.0 is

xAI calls Grok Voice 3.0 “the world’s most intelligent speech-to-speech model,” with latency under one second (Introducing Grok Voice 3.0). Starlink call centers handle 20,000 customer calls a day across multiple languages, and about 70% are resolved with no human in the loop (Introducing Grok Voice 3.0).

Benchmarks

On the Artificial Analysis Speech-to-Speech Index, Grok Voice 3.0 ranks #1 with a score of 86.9, against 81.3 for Grok Voice 2.0; Full Duplex Bench is 97% for how naturally it handles pauses, turn-taking, and interruptions; τ-Voice is 71.6%, comparable with GPT-Live-1 (Astra) at 40% lower latency; and Big Bench Audio is 98.5%, from Artificial Analysis on October 13, 2026, as reported by xAI (Introducing Grok Voice 3.0). On EVA-Bench from ServiceNow, it records the highest accuracy, at 74% (Introducing Grok Voice 3.0).

It powers Grok Bot’s voice mode

Grok Voice 3.0 also powers Grok Bot’s voice mode (Introducing Grok Voice 3.0). On September 17, 2026, the official @bot account posted “Grok Bot can talk now,” and voice launched on desktop and mobile (Official @bot post).

In Tesla vehicles, drivers use Grok Bot for bigger tasks like ordering a coffee or booking dinner, completely hands-free (Introducing Grok Voice 3.0). Grok Voice 3.0 handles millions of conversations a day in Tesla vehicles (Introducing Grok Voice 3.0).

Three ways to use voice in Grok Bot

Live voice chat

Choose “Start voice chat” when the composer is empty to talk with the Bot live (Message and collaborate). While you are on the line, the app shows “Voice chat” (Message and collaborate). On desktop, the sidebar can show “Voice chat in progress” (Message and collaborate). When the call ends, the conversation can include a “Voice chat” card (Message and collaborate). “Start voice chat” has no keybinding (Message and collaborate).

Dictation

Dictate into the composer with Cmd/Ctrl+D while the prompt is focused (Message and collaborate). On desktop, Cmd/Ctrl+D or “Start voice input” dictates (Grok Bot FAQ). On iPhone and Android, the control is “Start dictation” (Grok Bot FAQ).

Voice memos

A Bot can reply with a voice memo: “Play voice memo,” “Pause voice memo,” and expand the memo to read the transcript (Message and collaborate). While a Bot’s voice memo plays, the transcript highlights each word in sync with the audio, and clicking a word jumps playback there, as of the September 18 changelog (Grok Bot Changelog). Anyone with Grok Bot access can play a Bot’s voice memos without being asked to upgrade their plan, as of the September 22 changelog (Grok Bot Changelog).

Desktop and mobile

“Start voice chat” works on desktop, iPhone, and Android (Grok Bot FAQ). On mobile, while you are on the line, a Voice chat control stays under the header (Grok Bot for Mobile).

Settings that matter

Voice chats use the microphone you choose under Microphone in Settings, falling back to the default microphone if that microphone is unavailable, as of the September 22 changelog (Grok Bot Changelog). A Bot keeps the voice, speed, and language you set for it, instead of resetting them to defaults, as of the September 29 changelog (Grok Bot Changelog). Voice calls offer 0.5x speed, as of the October 11 changelog, v0.70.2 (Grok Bot Changelog). Voice input detects the language you speak, instead of assuming your app or system language, as of the August 21 changelog, v0.24.0 (Grok Bot Changelog).

Team Bots

Teammates can start a voice chat with a Team Bot, and the call stays in their own conversation with that Bot, as of the September 30 changelog, v0.65.0, on Teams (Grok Bot Changelog). For how this site describes a roster of Bots, see agent teams.

Hanging up mid-task

If you hang up a voice chat while a Bot is still working on your request, it finishes and posts the result in the chat, as of the September 22 changelog (Grok Bot Changelog).

For builders: Voice Agent API

The Voice Agent API is $0.08/min ($4.80/hr), the same price as Grok Voice 2.0 (Introducing Grok Voice 3.0). Agents using grok-voice-latest were switched over automatically, and no changes were needed (Introducing Grok Voice 3.0).

The speech-to-speech API runs full-duplex real-time conversations with sub-second latency, tool use, search, and multi-turn exchanges, over WebSocket at $0.08/min (Grok Voice API). It offers expressive voices across 25+ languages (Grok Voice API).

Quick checklist

1. Choose “Start voice chat” when the composer is empty to talk with the Bot live; that action has no keybinding (Message and collaborate).
2. “Start voice chat” works on desktop, iPhone, and Android (Grok Bot FAQ).
3. On desktop, dictate with Cmd/Ctrl+D while the prompt is focused, or use “Start voice input”; on iPhone and Android, use “Start dictation” (Message and collaborate, Grok Bot FAQ).
4. On desktop, the sidebar can show “Voice chat in progress” during the call; on mobile, a Voice chat control stays under the header (Message and collaborate, Grok Bot for Mobile).
5. Voice chats use the microphone you choose under Microphone in Settings, and fall back to the default microphone if that choice is unavailable (Grok Bot Changelog).
6. A Bot keeps the voice, speed, and language you set for it, and voice calls offer 0.5x speed (Grok Bot Changelog).
7. Voice input detects the language you speak, instead of assuming your app or system language (Grok Bot Changelog).
8. Use “Play voice memo” and “Pause voice memo,” and expand the memo to read the transcript; while it plays, the transcript highlights each word, and clicking a word jumps playback there (Message and collaborate, Grok Bot Changelog).
9. Anyone with Grok Bot access can play a Bot’s voice memos without being asked to upgrade their plan (Grok Bot Changelog).
10. A voice chat with a Team Bot stays in that teammate’s own conversation (Grok Bot Changelog).
11. If you hang up while a Bot is still working on your request, it finishes and posts the result in the chat (Grok Bot Changelog).
12. Agents using grok-voice-latest were switched over automatically, and the Voice Agent API is $0.08/min ($4.80/hr), the same price as Grok Voice 2.0 (Introducing Grok Voice 3.0). The speech-to-speech API runs over WebSocket at $0.08/min, with full-duplex real-time conversations, sub-second latency, tool use, search, multi-turn exchanges, and expressive voices across 25+ languages (Grok Voice API).