Models · 14 October 2026 · 1 min read
Grok Voice 3.0: xAI’s voice model for phone agents
Released on 14 October 2026: a voice model for phone agents, answering in under a second, at the same price as the version before.
By The Crewnema team
xAI released Grok Voice 3.0 on 14 October 2026. It is a speech-to-speech model: it listens and answers in voice, without turning speech into text and back. These are the points from xAI’s announcement.
What xAI announced
- A speech-to-speech model in xAI’s Voice Agent API; agents that use the latest alias moved to it automatically.
- Latency under one second, and first place on Artificial Analysis’s speech-to-speech index at launch, by xAI’s account.
- $0.08 a minute, the same as Grok Voice 2.0.
- In use in Starlink call centres across several languages, in Grok’s voice mode and in Tesla cars.
Our view
A voice agent answers the phone; it does not read an advert. For a small business the use is a shop that cannot pick up every call, and at about five dollars an hour of conversation the price is low enough to try. For video, a model built to answer quickly is the wrong tool: a voiceover is written once and should be read with care, not speed.
Side by side
In Crewnema
Grok Voice 3.0 is not on Crewnema’s roster, and as a model for live conversation it is not the kind we would add for voiceovers.
Sources
Make your first video
Crewnema is in private beta and lets businesses in a few at a time. Request access and we will write when yours is open.