xAI released Grok Voice 3.0 on Oct. 14. It is an AI that listens to people and talks back out loud in real time. xAI is offering it to businesses that want AI to pick up the phone, answer questions and sort out problems for customers. xAI calls it the world's most intelligent voice model and says it sets a new record for resolving customer support calls.
Businesses can use it now through xAI's API, a connection that lets developers plug a service into their own apps, at $0.08 a minute, or $4.80 an hour, the same price as the previous version, Grok Voice 2.0. Apps already set to use xAI's latest voice model were switched to the new one automatically.
An AI answering phones has to do two hard things at once: work out what a caller wants, and answer quickly enough that the conversation feels natural. xAI says the model's latency, the delay before it responds, is under one second. In rankings published by an outside testing firm, Grok Voice 3.0 started speaking 0.81 seconds after a question and had the highest overall score of the models given one, 86.9 out of 100, ahead of versions of OpenAI's GPT-Live-1 at 83.2 and Google's Gemini 3.8 Live at 82.6. That score weighs four tests equally: answering reasoning questions asked out loud, handling simulated support calls, which AI people prefer after blind side-by-side conversations, and whether it got the job done in those conversations.
It does not lead every test. On simulated support calls, which xAI says include varied accents, background noise and interruptions, it resolved 71.6% of the calls, against 74.5% for the GPT-Live-1 version. xAI calls those results comparable and points out that its model answers faster. On the reasoning test alone, three other models scored slightly higher.
xAI says Grok Voice 3.0 already handles millions of conversations a day in Tesla cars, where drivers can ask it to check a calendar, answer an email, order a coffee or book dinner without using their hands. It also says Starlink's call centers use Grok Voice to answer 20,000 calls a day in several languages, and that about 70% of them are resolved without a person stepping in.
xAI's claimed record for resolving support calls does not show up on the outside firm's support-call test, where OpenAI's model scored higher. xAI points instead to a test of business voice agents made by the company ServiceNow, where it says its model had the highest accuracy, 74%. xAI's real-world examples all come from Tesla and Starlink. It is not yet clear how well the model handles real customers for other businesses, or how often it gets things wrong on a live call.