Grok Voice 3.0

Grok Voice 3.0 is xAI’s model for live spoken conversations: an application sends speech and receives spoken replies. The 14 October 2026 release notes announce its API availability and mark Voice 2.0 as deprecated.

The latest-model alias now selects version 3.0

The versioned identifier is grok-voice-3.0. Requests using grok-voice-latest now select this version. The speech-to-speech guide explains the supported session controls, audio events and tool calls.

Existing applications use the same endpoint, parameters and event structure. xAI says pricing is unchanged from Voice 2.0. Developers should check the current price schedule and test their own conversation flows before moving traffic.

Test the complete conversation

A voice application combines the model with microphones, playback, network transport and any tools it can call. Test interruption handling, recognition in the intended environment and confirmation before consequential actions. Those checks establish how a particular application behaves.

Speech-to-speech models handle the conversation itself. Standalone transcription converts audio into text; speech synthesis converts supplied text into audio. The audio directory covers those other tasks.