Announced October 14, Grok Voice 3.0 replaces the previous 2.0 generation model behind the grok-voice-latest model alias. xAI kept the underlying WebSocket endpoint, session configuration parameters, and event handling protocols identical to the previous generation, allowing existing integrations to receive the new model without structural code changes.
Model Configuration and Version Pinning
While projects pointed at grok-voice-latest automatically transition to 3.0, developers can select grok-voice-3.0 explicitly to pin the model version. Grok Voice 2.0 is now marked as deprecated, though xAI has not provided a final shutdown date.
The API supports a configurable reasoning effort parameter. The parameter defaults to high and can optionally be set to none.
On the security side, documentation continues to specify that primary API keys should remain on server-side architecture. Systems establishing client connections, such as browser sessions or mobile apps, are directed to use ephemeral client tokens generated on the server.
Pricing Boundaries and Scope
Usage pricing remains unchanged from Grok Voice 2.0 at $0.08 per minute, which equals $4.80 per hour. The pricing table also lists a $0.004 charge per text input. Server-side tools carry separate invocation charges under the platform pricing schedule.
In its release announcement, the company stated that the underlying voice architecture powers millions of daily conversations across Tesla and SpaceX products.