Workflow Studio runs one text turn in a fresh Grok Voice 3 session and returns a transcript and WAV. Select PCM output over JSON and clear native tools before applying. Live microphone, SIP, interruption, and resumption controls apply to this playground's live sessions.

Grok Voice 3

Voice & speech playground

Hydrating page from local storage...

Explore Grok Voice 3 with reasoning, all native tool types, precise pronunciation, telephony codecs, binary audio, scripted speech, and resumable conversations.

Microphone access requires HTTPS or localhost and your permission. Audio is sent to the selected provider only when you submit a request or start a live session. Use headphones for live conversations.

Connection

Use the native SpaceXAI endpoint.

Endpoint

Model / deployment name

Authentication Mode

API Key

Advanced

Speech controls

Response language / locale (optional)

Choose a response language or locale, or select Custom to enter another provider-supported code. The selection is included in response instructions and input transcription hints. Automatic language detection leaves the response language to your instructions and the detected speech.

Voice

Choose a built-in voice, or select Custom to enter another documented voice ID. Grok also accepts an existing authorized custom voice ID.

Vocabulary hints (one per line)

Turn detection

Native web search Native X search

Grok Voice 3 is the current grok-voice-latest model. Use grok-voice-3.0 to pin it. Realtime usage costs $0.08/minute plus $0.004 per text input; tool usage can add charges. Voice library and SIP utilities

Reasoning effort

Output speech speed

Idle re-engagement (ms, optional)

Requires server VAD. Repeats after every response until you speak. Leave empty to disable.

Ephemeral token lifetime (seconds)

Enable microphone (turn off for text-only input) Live, revisable input captions (grok-transcribe) Enable server-side conversation resumption

Resumption opts in to provider-side caching of transcripts and tool results. History expires after 30 minutes of inactivity. The conversation ID is retained only in this page's memory.

Input codec

Input PCM sample rate

Input audio transport

Output codec

Output PCM sample rate

Output audio transport

Opus requires WebCodecs support. Unsupported codecs produce a visible error before connecting. Captured realtime audio is decoded and saved as WAV for playback and download.

Pronunciation replacements (JSON object)

Replacements change spoken output while keeping the original transcript. Up to 200 whole-word phrases. Use a respelling or IPA between forward slashes.

Additional native tools and function schemas

Supports web_search and x_search filters, file_search collections, multiple MCP servers, and custom functions. Server-side tools execute at xAI. Function calls wait for the JSON result you enter in Live session tools. Remote tools can read shared conversation context and perform actions; choose trusted servers and narrow allowed_tools.

Additional tools (JSON array)

[]

MCP authorization (optional, protected secret) MCP headers (optional JSON object, protected secret)

These secret fields apply to every configured MCP server. Opt-in auto-save encrypts them; non-secret profiles omit them.

System instructions

Ready. Configure your connection to begin.

Text message (optional)

Estimated cost Unavailable(Reported Cost: Unavailable)

xAI: $0.08/minute plus $0.004 per text input. Estimate uses observed connection time and sent text; native tool charges are excluded.

Transcripts and captured speech will appear here. Configure your endpoint, then record or start a live session.