Grok Voice 3.0 Explained: xAI's New Speech Model, Pricing & How to Try It
What the Grok Voice 3.0 speech model is, how xAI's speech-to-speech model scores against GPT-Live-1 and Gemini Live, what $0.08 per minute really costs, and how to try it — plus when you need text to speech instead.
Product note · Use AI-generated voices responsibly
Quick answer: Grok Voice 3.0 is xAI's speech-to-speech model, released on October 14, 2026. It hears a speaker, reasons, calls tools, and talks back in under a second, all in one model. It ranks #1 on Artificial Analysis's Speech-to-Speech Index (86.9) and costs $0.08 per minute in xAI's Voice Agent API, the same price as Grok Voice 2.0. It is built for live conversations such as support calls and in-car assistants. It is not a narration or voice-over tool.
Voiceshard is not affiliated with xAI. It does not offer Grok Voice 3.0 conversations, but it does offer xAI's separate Grok TTS voices in its Grok text to speech tool. Facts below come from xAI's announcement and developer docs, checked October 16, 2026.
What is the Grok Voice 3.0 speech model?
The Grok Voice 3.0 speech model works speech-to-speech: audio goes in and audio comes out, without a separate speech-to-text → language model → text-to-speech chain. In a single session it can listen, think, use tools (web search, X search, file search, MCP servers and your own functions), and answer out loud.
xAI says the model already runs in production:
- Tesla vehicles: millions of conversations a day, from calendar checks to ordering coffee hands-free.
- Starlink support: about 20,000 calls a day across languages; roughly 70% are resolved with no human, and one in four sales calls ends in a purchase.
- Grok Bot's voice mode also runs on Grok Voice 3.0.
These deployment figures are xAI's own and have not been independently audited.
For developers the model ID is grok-voice-3.0. The alias grok-voice-latest already points to it, so existing agents were upgraded automatically, and Grok Voice 2.0 is now deprecated. It uses the same endpoint, session parameters, and events as 2.0, and speaks 20+ languages.
Grok Voice 3.0 vs 2.0, GPT-Live-1 and Gemini Live
Benchmarks below are from Artificial Analysis (October 13, 2026) as published by xAI. Higher is better.
| Benchmark | Grok Voice 3.0 | Grok Voice 2.0 | GPT-Live-1 (Astra, Medium) | Gemini 3.8 Live (Ext. Thinking, High) |
|---|---|---|---|---|
| AA Speech-to-Speech Index | 86.9 | 81.3 | 83.2 | 82.6 |
| Big Bench Audio (reasoning) | 98.5% | 97.2% | 90.1% | 97.7% |
| Speech Agent Arena Elo (preference) | 1070 | 1011 | 1048 | 990 |
| Task success rate | 94.8% | 94.6% | 87.4% | 89.1% |
| Full Duplex Bench (turn-taking) | 97.0% | 95.1% | 94.9% | 91.9% |
Two more data points from the announcement:
- τ-Voice (support calls with accents, noise and interruptions): 71.6%, comparable with GPT-Live-1 (Astra) but with about 40% lower latency, according to xAI.
- EVA-Bench (ServiceNow's enterprise voice-agent benchmark): the highest accuracy score, at 74%.
Benchmarks measure agent tasks such as resolving a call. They do not tell you how a model sounds for narration or character voices.
Grok Voice 3.0 pricing: what $0.08 per minute costs
| Item | Price |
|---|---|
| Speech-to-speech (Grok Voice 3.0) | $0.08 per minute ($4.80 per hour) |
| Text input inside a session | $0.004 per text input |
Rough math at list price:
- One 5-minute call: $0.40
- 1,000 calls of 3 minutes each: $240
- An agent talking 8 hours a day for 30 days: about $1,152
Artificial Analysis also measured cost per completed task on its Big Bench Audio run. Grok Voice 3.0 came in at about $0.025, below GPT-Live-1 (Astra, Medium) and GPT-Realtime-2.1. Treat cost per task as a benchmark figure, not a bill: real costs depend on call length and how often the agent uses tools or text input.
How to try Grok Voice 3.0
- In Grok Bot's voice mode. xAI says Grok Bot's voice mode runs on Grok Voice 3.0. What you can access depends on your xAI product and plan.
- In the xAI console. The announcement links a "Try It Free" button to the Voice Agents page in the xAI developer console, where you can talk to an agent before writing code.
- Through the API. Developers build their own voice agents on the same model. See Grok Voice Agent API below.
Grok Voice Agent API: build a voice agent with Grok Voice 3.0
What a Grok voice agent can do
A Grok voice agent is an app that holds live spoken conversations through xAI's Voice Agent API, with Grok Voice 3.0 as the model. In one session it listens, reasons, calls tools (web search, X search, file search, MCP servers and your own functions) and answers out loud in under a second, in 20+ languages. You pay $0.08 per minute while the session is open, plus $0.004 per text input. Because it works speech to speech, it never needs a written script, which is also why it is not a narration tool.
How to connect to the Grok Voice API
Agents connect over WebSocket at wss://api.x.ai/v1/realtime with model=grok-voice-3.0, or with the alias grok-voice-latest, which now points to 3.0. Agents built on Grok Voice 2.0 keep the same endpoint, session parameters and events. Browser apps should not ship an API key: your server requests a short-lived client secret from /v1/realtime/client_secrets, and the browser passes it as a WebSocket subprotocol. Setup steps and code samples are in xAI's Voice API docs.
Grok Voice 3.0 vs text to speech: which one do you need?
They solve different jobs:
- Speech-to-speech (Grok Voice 3.0): a live, two-way conversation. You pay by the minute while the line is open.
- Text to speech: you write a script and get a finished audio file for videos, ads, games or social clips. xAI sells this separately in the same Voice API at $15 per million characters, through its own endpoint.
If you came here because you want a voice reading your script, you need text to speech, not a conversational agent. You can try free text to speech on Voiceshard with ready-made AI voices in the browser, or use AI voice cloning to make a private voice from your own recording. If you want those Grok TTS voices without writing code, Voiceshard offers them in its own tool: see Grok text to speech.
FAQ
Is Grok Voice 3.0 free?
The API is paid, at $0.08 per minute. xAI's announcement links a free way to try voice agents in its console, and Grok Bot's voice mode runs on the model. What consumer apps include depends on your plan, so check there.
How much does Grok Voice 3.0 cost, and how fast is it?
It costs $0.08 per minute ($4.80 per hour) in xAI's Voice Agent API, the same as Grok Voice 2.0, plus $0.004 per text input. xAI says it answers in under one second.
Is Grok Voice 3.0 a text-to-speech model?
No. It is a speech-to-speech model for live conversations. xAI offers text to speech as a separate endpoint in the same Voice API.
Can Grok Voice 3.0 clone a voice?
xAI's Voice API supports custom voices made from a reference clip of up to 120 seconds, usable in both speech-to-speech and text to speech. According to xAI's docs, custom voices are currently available only in the United States, excluding Illinois. Creating them through the API is limited to Enterprise teams; other teams create them in the console.
What languages does Grok Voice 3.0 support?
xAI lists 20+ languages, including English, Spanish, French, German, Hindi, Japanese, Korean, Simplified Chinese, Arabic and Portuguese.
What is grok-voice-latest?
It is the model alias in xAI's Voice Agent API that points to the newest Grok Voice model, now grok-voice-3.0. Agents that used the alias were switched to 3.0 automatically, with no code changes, and Grok Voice 2.0 is deprecated.
Does Voiceshard use Grok Voice?
No. Voiceshard does not offer Grok Voice 3.0 conversations and is not affiliated with xAI. It does offer xAI's separate Grok TTS voices in its Grok text to speech tool.
Sources
- Introducing Grok Voice 3.0 — xAI, October 14, 2026
- Voice API docs and pricing — xAI developer docs, checked October 16, 2026
- Speech-to-Speech leaderboard — Artificial Analysis