Grok Voice 3.0 leads Artificial Analysis’s composite voice index
Artificial Analysis gives Grok Voice 3.0 an index score of 86.9 across four equally weighted components.
In this article
What the score measures
The benchmark combines reasoning from spoken questions, simulated customer-support performance, human preference and task success. Only models with all four results receive an index score. Conversation flow and response speed have separate measures; they are not components of this composite.
MarkTechPost’s coverage also discusses the release, linking to the announcement and benchmark.
Availability and existing applications
In its October 14, 2026 announcement, SpaceXAI said the model was available that day through its Voice Agent API, the interface applications use to access the service. The provider listed a price of $0.08 per minute, unchanged from the previous generation. It also said agents using the latest-model alias had already switched automatically, without requiring changes. These are the provider’s availability and migration statements.
Choosing the model connection
The API guide describes a WebSocket connection that carries audio and text in both directions for voice assistants and phone agents. The model query parameter selects the release: grok-voice-latest currently points to grok-voice-3.0, while a versioned name pins a specific release. This is model selection, not a claim that repeated evaluations produce identical results. For authentication, the guide recommends short-lived tokens in browser or mobile applications and restricts direct API-key authentication to servers.
Sources(4)
- Release coverageChecked:
- Launch noticeChecked:
- Evaluation tableChecked:
- Connection guideChecked:
