T9-R6 in five minutes
The Speech to Speech page says you connect with a WebSocket to wss://api.x.ai/v1/realtime and pass model as a query parameter.
That page says grok-voice-latest is an alias for grok-voice-3.0, and that grok-voice-2.0 is deprecated.
Session fields on that page include instructions, voice, tools, and turn_detection. The tools it lists are file_search, web_search, x_search, mcp, and function.
Authentication on that page is an ephemeral token for a client-side app, or an API key in the Authorization header for the server side only.
The Voice overview says all audio data is processed in real time and never stored or used for training. That is the page's statement. This lab sends no audio.
Quote the page you opened. The quick-start sample uses the voice name eve and server_vad. Those are sample fields, not a contact-center standard. This page states no price.
Sources and go deeper
- Speech to Speech, docs.x.ai vendor The page title is Speech to Speech. The body names the Grok Speech to Speech API and also calls the site SpaceXSI Docs. last checked 2026-10
- Voice Overview, docs.x.ai vendor The page calls them SpaceXSI Voice APIs, powered by Grok, and separates Speech to Speech, Text to Speech, and Speech to Text. last checked 2026-10