Grok Voice 3.0 launches API access at $0.08 per minute

• Grok Voice 3.0 is available through the API and Voice Agent Builder for $0.08 per minute, per the availability announcement.
• Voice replies arrive in under a second, according to the launch announcement.
• About 70% of Starlink support calls are resolved without a human, according to the company's deployment figures.

The Builder offers a neat brand-design shortcut: voice cloning from two minutes of audio. Existing agents using grok-voice-latest have already been upgraded automatically.

SpaceXSI released Grok Voice 3.0 on October 14, keeping its predecessor's price of $0.08 per minute, or $4.80 per hour, according to the launch post.

Custom apps connect over WebSockets at wss://api.x.ai/v1/realtime?model=grok-voice-latest. The API documentation describes tool use and ephemeral tokens for client-side connections without exposing a permanent API key.

A short conversation demo appears in minchoi's launch post:

Grok Voice 3.0 conversation demo, shared by minchoi

The no-code Builder debuted in beta in July, ahead of this model release. Its current configuration options include:

• Instructions: a playbook covering the greeting, resolution steps and wrap-up.
• Knowledge: uploaded PDFs, Markdown and plain-text documents.
• Actions: connected apps such as Gmail and Google Calendar, custom APIs and MCP servers.
• Testing: live browser conversations and recordings for call playback.
• Guardrails: configurable limits on what the agent can do.
• Phone deployment: a free phone number or an existing number through SIP, plus tools for handing calls to a human.

Grok Voice 3.0 ranks first on Artificial Analysis's composite Speech-to-Speech Index. Compared with Voice 2.0:

• Overall index: 81.3 → 86.9, up 5.6 points.
• τ-Voice support-call performance: 56.5% → 71.6%, up 15.1 percentage points.
• Full Duplex Bench, covering pauses, turn-taking and interruptions: 95.1% → 97.0%, up 1.9 percentage points.

First audio takes 0.81 seconds versus 1.34 seconds for GPT-Live-1 (Astra, medium), consistent with the company's 40% lower-latency claim. Voice 2.0 was quicker at 0.70 seconds in the same independent table, so the new model adds 110 milliseconds against its predecessor.

At Tesla, the company reports millions of daily conversations, including calendar checks through Connectors and coffee orders through Grok Bot, according to the launch post. The model also powers Grok Bot's voice mode.

Starlink's deployment handles about 20,000 customer calls daily across multiple languages. The company says one agent can verify customers, troubleshoot hardware, issue credits and ship replacements; one in four sales calls ends in a purchase before the caller hangs up.