ModelGrok

xAI launches Grok Voice 3.0: under 1 second response

Original title: Introducing Grok Voice 3.0

Plain language

What it is: Grok announced Grok Voice 3.0, a new speech-to-speech model for voice conversations. It is designed to resolve customer support calls quickly, with a response time of under one second, and costs less than comparable models.
How it works: The model handles natural conversation flow, including pauses, interruptions, and taking turns speaking. It can also understand different accents and handle background noise during calls.
Where it is available: It is available in the Speech to Speech API. It is also used in Tesla vehicles and Starlink call centers.
Example: Starlink call centers use the model to answer customer calls, verify customer identities, troubleshoot hardware issues, and issue credits without human intervention.

What it is: Grok Voice 3.0 is a speech-to-speech model that ranks first on Artificial Analysis's Speech-to-Speech Index with a score of 86.9. It features sub-second latency and lower operating costs compared to other frontier models.
How it works: The model processes speech-to-speech directly. It scores 97% on the Full Duplex Bench for handling pauses and interruptions, and 71.6% on the τ-Voice metric for handling diverse accents and background noise.
Where it is available: It is available via the Speech to Speech API at $0.08 per minute. Existing integrations using the grok-voice-latest alias have been updated automatically.
Example: Developers can deploy the model to build automated customer service agents that verify users, troubleshoot hardware, and process equipment shipments.

The headline and explanation on this page are based on xAI's post. The full details are in the original.

Introducing Grok Voice 3.0