Grok Voice 3.0: Features, API Guide, Pricing and Reviews

Grok Voice 3.0: Features, API Guide, Pricing and Reviews

AI Takeaway Teamon 2026-10-16

Last verified: October 16, 2026. Facts come from xAI's (now SpaceXSI) official pages and docs linked below. Benchmark and customer figures are vendor-reported unless marked as independent. We have not run our own tests.

Grok Voice 3.0 is xAI's newest speech-to-speech model for real-time voice agents. It was announced on October 14, 2026 in Introducing Grok Voice 3.0 as "the world's most intelligent speech-to-speech model." The official @SpaceXSI announcement says it "excels at resolving customer calls, responds in under a second, and costs less than other frontier models." xAI now appears under the SpaceXSI name on its site and docs; this guide uses both names. The launch was posted to Hacker News. Below: what is new, how it differs from Grok Voice 2.0, how to call it through the API, and what independent testers and users report.

At a glanceGrok Voice 3.0
What it isSpeech-to-speech model for real-time voice agents
API model IDgrok-voice-3.0 (alias: grok-voice-latest)
Price$0.08 per minute ($4.80 per hour), same as Grok Voice 2.0
Endpointwss://api.x.ai/v1/realtime (WebSocket)
Regions (model page)us-east-1, eu-west-1, us-saltlake-2
Previous versionGrok Voice 2.0, now deprecated

Sources: xAI announcement, Grok Voice 3.0 model page, Speech to Speech docs.

Practical rule: if your agent already uses grok-voice-latest, it is on Grok Voice 3.0 today. If you pinned grok-voice-2.0, change one string before it stops working.

What's new in Grok Voice 3.0

xAI's launch post highlights four things:

  • Accuracy: it ranks #1 on Artificial Analysis's Speech-to-Speech Index with a score of 86.9, and scores 97% on Full Duplex Bench, a measure of "how naturally it handles pauses, turn-taking, and interruptions."
  • Speed: "class-leading latency at under one second."
  • Price: "the same price as its predecessor."
  • Real-world use: xAI says Grok Voice handles "millions of conversations a day in Tesla vehicles," and that Starlink call centers use it "to answer 20,000 customer calls every day across multiple languages," with "roughly 70%" resolved with no human in the loop. These usage figures are xAI's own claims.

The docs also note that reasoning is enabled by default for this model: the default reasoning.effort is "high", and it can be set to "none" to turn reasoning off. They advise simplifying system prompts when migrating, because "the model is significantly more capable."

Grok Voice 3.0 vs Grok Voice 2.0

The comparison below is xAI's table, which uses Artificial Analysis data from October 13, 2026 (so it is third-party-measured but published by xAI):

BenchmarkGrok Voice 3.0Grok Voice 2.0GPT-Live-1 (Astra, Medium)Gemini 3.8 Live (Ext. Thinking, High)
Overall AA Speech-to-Speech Index86.981.383.282.6
Speech reasoning (Big Bench Audio)98.5%97.2%90.1%97.7%
Human preference (Speech Agent Arena Elo)107010111048990
Task completion (success rate)94.8%94.6%87.4%89.1%
Conversational dynamics (Full Duplex Bench)97.0%95.1%94.9%91.9%

The biggest jumps over 2.0 are in the overall index (+5.6) and human preference (+59 Elo). Task completion barely moved (94.8% vs 94.6%). xAI adds that on τ-Voice, Grok Voice 3.0 scores 71.6%, "comparable with GPT-Live-1 (Astra), but at 40% lower latency," and that on ServiceNow's EVA-Bench it "achieves the highest accuracy score at 74%." Independent coverage from SiliconANGLE reports the same 86.9 index score.

On the API side, nothing changes: "Grok Voice 3.0 uses the same endpoint, session parameters, and events as Grok Voice 2.0, at the same price" (docs).

Vendor-published tables are a starting point. Test with your own audio, accents and phone lines.

How to use the Grok Voice 3.0 API

The Voice Agent / Speech to Speech API runs over a WebSocket at wss://api.x.ai/v1/realtime. You pass the model as a query parameter. The docs list three values: grok-voice-latest (alias for grok-voice-3.0), grok-voice-3.0, and grok-voice-2.0 (deprecated). They also advise: "Pin to a versioned model name in production for stability."

Authenticate with an API key in the Authorization header (server-side only), or use short-lived ephemeral tokens for browsers and mobile apps. Here is the official Python quick start from the Speech to Speech docs, with the model pinned to grok-voice-3.0:

import asyncio
import json
import os
import websockets

async def voice_agent():
    async with websockets.connect(
        "wss://api.x.ai/v1/realtime?model=grok-voice-3.0",
        additional_headers={"Authorization": f"Bearer {os.environ['XAI_API_KEY']}"}
    ) as ws:
        # Configure session
        await ws.send(json.dumps({
            "type": "session.update",
            "session": {
                "voice": "eve",
                "instructions": "You are a helpful assistant.",
                "turn_detection": {"type": "server_vad"}
            }
        }))

        # Send a text message
        await ws.send(json.dumps({
            "type": "conversation.item.create",
            "item": {"type": "message", "role": "user",
                     "content": [{"type": "input_text", "text": "Hello!"}]}
        }))
        await ws.send(json.dumps({"type": "response.create"}))

        # Receive audio/text responses
        async for msg in ws:
            event = json.loads(msg)
            print(f"Event: {event['type']}")

asyncio.run(voice_agent())

Useful options documented for session.update:

  • Tools: file_search, web_search, x_search, mcp and custom function tools.
  • Audio formats: audio/pcm, audio/pcmu, audio/pcma and audio/opus.
  • Languages: 20+ languages, with automatic detection of the input language.
  • Custom voices: clone a voice from a short audio clip.
  • Session resumption: replay prior context on reconnect; history is dropped after 30 minutes of inactivity.
  • Pronunciation fixes: use replace to fix how brand names are spoken.

To upgrade from 2.0, set model to grok-voice-3.0 or grok-voice-latest. If you are moving from the OpenAI Realtime API, the docs say it "requires only a few changes: update the base URL, swap your API key, and choose a Grok voice model." The release notes confirm: "grok-voice-latest now points to grok-voice-3.0... The API is unchanged" (release notes).

Pricing

Grok Voice 3.0 costs "$0.08 / min ($4.80 / hr), the same price as Grok Voice 2.0" (xAI). xAI also says it is cheaper to run than comparable models on Artificial Analysis's Big Bench Audio run, under both GPT-Live-1 (Astra, Medium) and GPT-Realtime-2.1. The article gives that chart without exact figures, so we don't quote cost numbers for rivals.

Real-world reviews and early reactions

Early reactions are mostly positive but not unanimous. These are single-user posts, not controlled tests:

  • David Tran: "Loving the current Grok Voice 3, just used it walking and driving and it worked excellently with no mistakes so far." (post)
  • Marlon, who built an AI receptionist for a client: "Cost to run: 8 cents a minute! lol" (post)
  • Giuliano: "Was about to use Grok Voice 2.0 for UI control & consumer UX improvement. Grok Voice 3 is game." (post)
  • Rogu: "Grok Voice 3 is extremely bad and unpleasant to talk to." (post)
  • Faisal: "GPT-live-1 >>>> Grok Voice 3" (post)

On Hacker News, the launch thread is small (6 points at the time of writing). One commenter wrote that "hackernews is sleeping on this model" (comment).

Watch on YouTube: "Grok Voice 3.0 Review: Can It Get My Coffee Order Right?" by Cynxio

Third-party video by Cynxio, an independent creator. AI Takeaway did not make this video and is not affiliated with its creator.

No time to watch the whole review? AI Takeaway is a tool that works on any YouTube video, including third-party reviews like this one. Paste the link to get the key points in minutes, jump to the original timestamps, and ask follow-up questions; every answer cites the video. You can also browse the Hub for more AI and tech interviews.

Who should use Grok Voice 3.0

  • Customer-support and phone agents: the main target, per xAI. Native G.711 μ-law/A-law support and SIP phone-call routing are listed in the docs.
  • Existing Grok Voice 2.0 users: the upgrade is a one-line change at the same price.
  • Teams on the OpenAI Realtime API: xAI documents a short migration path, but compare quality on your own calls first, since early user opinions differ.
  • Probably not for: anyone who needs published latency numbers per region, or who cannot send audio to a hosted API.

How it compares (briefly)

On xAI's Artificial Analysis table, Grok Voice 3.0 scores 86.9 against 83.2 for OpenAI's GPT-Live-1 (with Astra reasoning) and 82.6 for Google's Gemini 3.8 Live. For the text-model side of the same race, see our guides to GPT-6.1 Sol vs GPT-6 Astra, Gemini 4 Argon and Meta's Muse Wonder.

Getting started checklist

  1. Create an xAI API key, and use ephemeral tokens for any browser or mobile client.
  2. Connect to wss://api.x.ai/v1/realtime?model=grok-voice-3.0 and send a session.update.
  3. Pin grok-voice-3.0 in production; use grok-voice-latest only if you want automatic upgrades.
  4. Test 50–100 real calls with your accents, noise and tools, and compare cost per resolved call.
  5. Re-check the docs before launch.

Keep up without the watch time

Voice models, frontier models and agent launches arrive weekly. If you would rather read than watch, the AI Takeaway video summary hub has free curated summaries of AI and tech interviews, for example:

To keep up with ongoing AI news, check the Hub regularly. You can also install the AI Takeaway Chrome extension to get key points right on YouTube, or submit any YouTube video yourself on aitakeaway.net.

For more model guides and comparisons, browse all AI Takeaway articles.

FAQ

What is Grok Voice 3.0?

A speech-to-speech model for real-time voice agents from xAI (SpaceXSI), announced October 14, 2026. The API model name is grok-voice-3.0.

How much does Grok Voice 3.0 cost?

$0.08 per minute ($4.80 per hour), the same as Grok Voice 2.0.

What's the difference between Grok Voice 3.0 and 2.0?

On xAI's table, 3.0 scores 86.9 vs 81.3 on the Artificial Analysis Speech-to-Speech Index. The API is unchanged and the price is the same. Grok Voice 2.0 is deprecated.

Do I need to change my code to use Grok Voice 3.0?

Not if you use grok-voice-latest, which already points to grok-voice-3.0. If you pinned grok-voice-2.0, set model to grok-voice-3.0.

Which API endpoint does it use?

wss://api.x.ai/v1/realtime, with the model passed as a model query parameter.

Sources