Grok Voice 3.0 launches with unchanged API pricing and auto-migration

xAI has released Grok Voice 3.0, making the speech-to-speech model available in its Voice Agent API while keeping pricing unchanged from Grok Voice 2.0, according to the company’s Oct. 14, 2026 announcement and the same-day API release notes.

Evidence and context

In its announcement, xAI described Grok Voice 3.0 as its newest voice model and said it is available today through the API at $0.08 per minute, or $4.80 per hour, the same price as the prior version. The company also said developers using grok-voice-latest have been switched over automatically, and the Oct. 14 release notes confirm that Grok Voice 2.0 has been deprecated with migration directed to Voice 3.0.

The release notes verified as of Oct. 14, 2026 identify Grok Voice 3.0 as the current flagship speech-to-speech model, with real-time bidirectional audio and text streaming over WebSocket for assistants, phone agents and interactive voice systems. Existing integrations retain the same endpoint, session parameters and events, while client-side applications can authenticate with short-lived ephemeral tokens instead of exposing API keys. The verified notes establish availability, migration behavior and unchanged pricing, but do not independently confirm the performance claims in xAI’s launch post.

xAI said Grok Voice 3.0 powers millions of conversations a day across Tesla and SpaceX. The company said Tesla vehicles use the model for hands-free tasks through Connectors and Grok Bot voice mode, while Starlink call centers handle 20,000 customer calls per day across multiple languages, with roughly 70% resolved without a human in the loop and one in four sales calls ending in a purchase before hang-up. Those operational figures were reported by xAI in its launch material and were not independently verified in the supplied material.

On performance, xAI cited third-party and benchmark results from Artificial Analysis and ServiceNow. The company said Grok Voice 3.0 ranked first on Artificial Analysis’s Speech-to-Speech Index with a score of 86.9, compared with 81.3 for Grok Voice 2.0, 83.2 for GPT-Live-1 (Astra, Medium) and 82.6 for Gemini 3.8 Live (Ext. Thinking, High). It also said the model scored 97.0% on Full Duplex Bench, 98.5% on Big Bench Audio speech reasoning, 94.8% on task success, and 1,070 Elo on Speech Agent Arena. In ServiceNow’s EVA-Bench, xAI said the model reached 74% accuracy and ranked best when accuracy and customer experience were scored together. These benchmark results are company-cited figures from the launch material and should be read in that context.

The broader Grok Voice platform, according to xAI’s product pages, supports full-duplex conversations, interruption handling, tool calling, search, multilingual speech in more than 25 languages, expressive voices and custom voice cloning. The supplied material distinguishes those platform capabilities from version-specific changes in Grok Voice 3.0.