xAI launches Grok Voice 3
xAI said the speech-to-speech model handles millions of conversations a day in Tesla vehicles and at Starlink's call centres, at under one second of latency.
- Models & capabilities
- Minor
xAI released Grok Voice 3.0, a speech-to-speech model the company described as its most intelligent voice system yet, with latency under one second and a lower running cost than its predecessor. It replaced Grok Voice 2.0 in the Voice Agent API at the same price, and xAI said agents already pointed at grok-voice-latest were switched over automatically.
xAI cited independent testing by Artificial Analysis putting Grok Voice 3.0 first on that firm’s Speech-to-Speech Index, ahead of OpenAI’s and Google’s comparable voice models, and reported strong scores on a benchmark of turn-taking and interruption handling. ServiceNow’s Nitin Aggarwal was quoted saying the model set “a new Pareto frontier” on the company’s own enterprise voice-agent benchmark, EVA-Bench.
The model already has large-scale deployment through xAI’s corporate ties to Elon Musk’s other companies: xAI said it handles millions of conversations a day inside Tesla vehicles, powering hands-free tasks such as checking a calendar or booking a restaurant, and that Starlink call centres use it to answer around 20,000 customer calls daily across multiple languages, with roughly 70% resolved without a human agent. Those figures came from xAI’s own announcement rather than an independent audit. The release illustrates how voice interfaces — rather than text chat — have become a competitive front between the frontier labs, with OpenAI’s and Google’s live voice models the explicit points of comparison in xAI’s own benchmark claims.