Grok Voice 3.0: xAI’s voice model with under a second of latency
xAI released a new speech model for voice agents at $0.08 a minute. Starlink already handles about 20,000 calls a day with it.
xAI has released Grok Voice 3.0, a speech-to-speech model aimed mainly at voice agents and customer support.
What’s new
- Latency: under one second, and about 40% lower than GPT-Live-1 at comparable accuracy, according to xAI.
- Conversation: it handles pauses, turn-taking and interruptions better (97% on Full Duplex Bench).
- Ranking: first on the Artificial Analysis Speech-to-Speech Index at 86.9, up from 81.3 for the previous version.
It doesn’t lead everywhere. On τ-Voice, a test of support calls with accents and background noise, it scored 71.6%, behind GPT-Live-1 at 74.5%.
Where it’s running
The model powers Grok Bot’s voice mode and Tesla vehicles. According to xAI, Starlink call centers use it to answer about 20,000 calls a day and resolve roughly 70% with no human involved.
How to try it
It’s available in the Speech to Speech API at $0.08 a minute ($4.80 an hour), the same as the previous version. Agents on grok-voice-latest were switched over automatically.
Limitations
Most of the numbers come from the company. The announcement doesn’t list supported languages, so test yours separately.
Prepared with AI from public sources.