Grok Voice 3.0: API Pricing and Capabilities

Grok Voice 3.0 is xAI’s speech-to-speech model for real-time voice agents, not a separately documented consumer app feature with its own download. As of October 17, 2026, xAI lists it as available through the Voice Agent API at $0.08 per minute, or $4.80 per hour; agents using grok-voice-latest were switched automatically.

What Is Grok Voice 3.0?

Grok Voice 3.0 converts spoken input into spoken responses while handling interruptions, pauses, turn-taking, and business actions. It is designed mainly for developers building voice agents, customer-support phone systems, browser experiences, mobile apps, and other real-time voice products.

xAI announced the model on October 14, 2026. Its primary published use cases include:

Voice agents that answer customer calls
Tesla vehicle conversations and hands-free requests
Starlink customer support
Agents that verify customers, troubleshoot hardware, issue account credits, and ship replacement equipment
Voice interfaces that connect to external tools and business systems

This is different from ordinary text chat with text-to-speech added afterward. A speech-to-speech model is intended to preserve conversational timing and respond naturally when a speaker pauses, interrupts, or changes direction.

What Changed in Version 3.0?

According to xAI’s announcement, Grok Voice 3.0 received the following results in Artificial Analysis testing:

86.9 on the Speech-to-Speech Index
98.5% on Big Bench Audio speech reasoning
94.8% task success rate in Speech Agent Arena
97.0% on the Full Duplex Bench
71.6% on the τ-Voice test for support-call conditions

The Full Duplex Bench measures conversational behavior such as pauses, turn-taking, and interruptions. The τ-Voice test considers accents, background noise, and interruptions in support calls.

xAI also reports response latency of under one second and says latency is about 40% lower than comparable systems in the relevant evaluation. These figures describe model and benchmark performance, not a guarantee that every phone call or device will respond in under one second. Internet connection, audio hardware, tools, and your application code can all add delay.

On ServiceNow’s EVA-Bench for enterprise voice agents, xAI reports a 74% accuracy score. Benchmark results are useful for comparison, but they do not prove that the model will perform equally well in your industry or call flow.

How to Use Grok Voice 3.0

For developers using the API

Create or access an xAI API account.
Create an API key in the xAI developer console.
Connect your application to the Voice Agent or real-time speech-to-speech interface.
Select the model name grok-voice-3.0 for a fixed production version.
Configure the agent’s instructions, voice, tools, and turn detection.
Stream audio into the session and play the returned audio to the user.
Test interruptions, silence, background noise, tool calls, and escalation to a human.

The published integration examples use a real-time WebSocket endpoint with the grok-voice-3.0 model. Browser and mobile applications should use short-lived or ephemeral tokens rather than exposing a permanent API key in client-side code.

A session configuration can include a voice, instructions, and server-side voice activity detection. In practice, your instructions should explain the agent’s role, what it may do, when it must verify identity, and when it must transfer the conversation.

If you already use `grok-voice-latest`

xAI says agents using grok-voice-latest were automatically moved to Grok Voice 3.0 without code changes. That is convenient for testing, but production applications generally benefit from a versioned model name so a future model update does not unexpectedly change prompts, timing, or tool behavior.

Before pinning or migrating, run calls through your real workflows. Test the calls that:

Ask for account or payment information
Use external tools
Require caller verification
Include frequent interruptions
Need a human handoff
Contain accents or background noise

Where Is Grok Voice 3.0 in the App?

The available research confirms an API release, not a separate “Grok Voice 3.0” switch in every Grok consumer app. The model powers Grok Bot’s voice mode according to xAI’s published announcement, but the exact menu location, account requirements, supported mobile versions, and country rollout for consumer voice mode are not specified in the announcement.

If you are trying to use voice in the Grok app:

Update the app from your device’s official app store.
Sign in to your Grok account.
Open a conversation and look for the voice or headphones control.
Allow microphone access when prompted.
Try a new conversation after updating.
Check xAI’s official product information at x.ai/news/grok-voice-3.

If the voice control is missing, do not assume that installing a third-party “Grok Voice 3.0” app will enable it. Availability may depend on the app version, account, platform, region, or staged rollout. The public API announcement does not establish universal consumer access.

Pricing and Plan Requirements

The published API price is $0.08 per minute, equal to $4.80 per hour. xAI says this is the same price as Grok Voice 2.0.

That price is for API usage. It does not confirm that consumer Grok voice mode is free, included with a particular subscription, or available under the same limits. App subscription pricing and API billing should be treated as separate products unless xAI explicitly connects them.

For a voice-agent project, estimate cost from actual call duration:

estimated cost = total voice minutes × $0.08

Your total project cost may also include telephony, hosting, storage, transcription, tool services, and human support.

Troubleshooting Voice Mode

If Grok voice does not work, check these items in order:

Microphone permission: Confirm that the Grok app or browser can access the correct microphone.
Audio output: Disconnect Bluetooth headphones temporarily and test the device speaker.
App version: Install the latest official app update.
Network: Switch between Wi-Fi and cellular data to test for connection problems.
Account or region: A missing voice control may reflect availability limits rather than a device failure.
Browser security: For a web implementation, use HTTPS and request microphone access from the correct origin.
API key handling: Never place a permanent xAI API key in a browser or mobile app.
Turn detection: If the agent cuts users off, adjust voice activity detection and silence thresholds.
Fallback behavior: Add a text option or human handoff when audio quality or tool execution fails.

For API debugging, log connection errors, session events, tool-call failures, and audio format details without storing sensitive call content unnecessarily.

FAQ

Is Grok Voice 3.0 available now?

Yes. xAI announced it as available through the Voice Agent API on October 14, 2026. Consumer app access is less clearly documented and may depend on platform, account, and region.

Is Grok Voice 3.0 free?

The API is priced at $0.08 per minute. The research does not confirm that the consumer app’s voice mode is free or included in a specific Grok subscription.

What devices support Grok Voice 3.0?

The official announcement describes API access, Tesla vehicle use, and voice-agent applications. It does not provide a complete consumer device compatibility list. For app access, use an officially supported current version of the Grok app and verify microphone permissions.

Does Grok Voice 3.0 support multiple languages?

xAI describes Starlink voice support as operating across multiple languages, but the announcement does not provide a complete supported-language list for every API or consumer app use case. Confirm language support for your specific integration before deployment.

Should I use `grok-voice-3.0` or `grok-voice-latest`?

Use grok-voice-3.0 when you need predictable production behavior. grok-voice-latest was automatically moved to version 3.0, but an alias can move again when xAI releases a later model.