For Developers/Models/Grok Voice 3.0
Closed SourcexAI (SpaceXAI)Released 2026-10

Grok Voice 3.0

xAI's speech-to-speech model for voice agents, ranked #1 on Artificial Analysis's speech index at launch

Context window

Not disclosed

Input / 1M tokens

Not applicable

Output / 1M tokens

Not applicable

Provider

xAI (SpaceXAI)

Billed per minute, not per token: $0.08 per minute ($4.80 per hour) of speech-to-speech audio plus $0.004 per text input, the same price as Grok Voice 2.0, per xAI's pricing page. Grok Voice 2.0 is now deprecated. · Data verified 2026-11-08

Grok Voice 3.0 (October 14, 2026) is xAI's speech-to-speech model for real-time voice agents, served through the xAI API over a realtime WebSocket. It replaces Grok Voice 2.0 at the same $0.08 per minute, and agents using the grok-voice-latest model name were switched to it automatically. Artificial Analysis ranked it first on its Speech to Speech Index at launch with 86.9, up from 81.3 for Grok Voice 2.0. xAI says it keeps latency under one second, handles pauses, interruptions, accents and background noise, and already runs Starlink support calls and voice conversations in Tesla vehicles.

How to access it

xAI API: connect to the realtime WebSocket endpoint wss://api.x.ai/v1/realtime with model grok-voice-3.0, or grok-voice-latest, which now points to it. Use ephemeral tokens for browser or mobile clients so the API key stays on your server.

Strengths

  • ✓86.9 on the Artificial Analysis Speech to Speech Index, ranked #1 at launch
  • ✓98.5% on Big Bench Audio speech reasoning, up from 97.2% for Grok Voice 2.0
  • ✓Flat $0.08 per minute, unchanged from Grok Voice 2.0
  • ✓Tool use and server-side voice activity detection in the realtime API
  • ✓Custom voices cloned from a reference clip of up to 120 seconds

Best for developers who...

Customer-support voice agentsPhone and call-center automationIn-app voice assistants that call tools

When to choose it (and when not to)

Reach for Grok Voice 3.0 when...

  • →You are building a voice agent and want the top-ranked speech-to-speech model at launch
  • →You want simple per-minute pricing
  • →Your agent needs tool calls during live calls

Look elsewhere if...

  • ✕You need the lowest possible latency: Artificial Analysis measured it slightly slower than Grok Voice 2.0 (0.81s vs 0.70s)
  • ✕You need self-hosting or open weights
  • ✕You need a text model for long documents or coding (use Grok 4.7)

How to use it

  • ›Pin grok-voice-3.0 if you do not want future model changes; grok-voice-latest moves automatically
  • ›Send a session.update message first to set the voice, instructions, turn detection and tools
  • ›Use server-side voice activity detection (server_vad) so the model handles turn-taking

Quickstart

Python
import asyncio, json, os
import websockets

async def main():
    url = "wss://api.x.ai/v1/realtime?model=grok-voice-3.0"
    headers = {"Authorization": f"Bearer {os.environ['XAI_API_KEY']}"}
    async with websockets.connect(url, additional_headers=headers) as ws:
        await ws.send(json.dumps({
            "type": "session.update",
            "session": {
                "voice": "eve",
                "instructions": "You are a friendly customer support agent.",
                "turn_detection": {"type": "server_vad"},
                "tools": [{"type": "web_search"}],
            },
        }))
        async for message in ws:
            event = json.loads(message)
            if event["type"] == "response.output_audio.delta":
                pass  # base64 audio chunk: decode and play it

asyncio.run(main())

Connection sketch adapted from xAI's voice guide, which uses grok-voice-latest; see the guide for streaming microphone audio.

API model id: grok-voice-3.0

Benchmarks

BenchmarkScoreNotes
Artificial Analysis Speech to Speech Index86.9Third-party, #1 at launch; Grok Voice 2.0 81.3
Big Bench Audio98.5%Speech reasoning; Grok Voice 2.0 97.2%
Full Duplex Bench97.0%xAI-reported conversational dynamics; Grok Voice 2.0 95.1%
Tau-Voice71.6%xAI-reported support calls with accents, noise and interruptions; xAI calls it comparable to GPT-Live-1
Speech Agent Arena Elo1070xAI-reported human preference; Grok Voice 2.0 1011

Source: xAI: Introducing Grok Voice 3.0

Compare Grok Voice 3.0

Compare Grok Voice 3.0 with any other model

Build a comparison →
All model comparisons →

Learn the concepts