Applications using grok-voice-latest have already moved to Grok Voice 3.0, according to xAI’s launch announcement. For teams using that floating alias, the release changes their production model without an explicit upgrade request.
Announced on October 14, 2026, Grok Voice 3.0 is available immediately through the Voice Agent API at $0.08 per minute, unchanged from Grok Voice 2.0. xAI claims latency below one second and reports higher benchmark scores than its predecessor, citing independent testing by Artificial Analysis.
For teams assessing the release, three kinds of evidence need separate consideration: published API terms, benchmark results attributed to an outside evaluator, and xAI’s own production claims. Together, they help frame what changed with the alias switch, but leave questions about how the model will perform in a team’s actual conversations.
The Price Stays Fixed, but the Model Behind the Alias Changes
The rate remains $0.08 per minute, or $4.80 per hour. At that listed rate, 10,000 billed minutes would cost $800 for the voice API alone.
A complete operating budget also needs to account for any separate telephony, infrastructure, external services, and application costs. Cost per completed task can change even when the per-minute rate stays fixed: a model may take more or fewer conversational turns to resolve an issue.
xAI’s voice API documentation lists grok-voice-latest as the speech-to-speech model, with a /v1/realtime endpoint and WebSocket-based conversations supporting tool use. This conversational API is distinct from the separately listed text-to-speech and speech-to-text services.
For existing alias users, xAI says the transition required no changes. That describes integration continuity, not a promise of identical behavior. The same connection code can now reach a model that follows instructions, handles interruptions, or chooses tools differently.
Agents authorized to take actions deserve particular scrutiny. A technically compatible replacement still needs behavioral validation before teams treat the migration as complete.
Sub-Second Latency Is a Vendor Claim, Not an Application Guarantee
xAI describes Grok Voice 3.0 as having latency below one second. Its launch material also presents a performance-versus-speed comparison using time to first audio, a measure of how quickly a spoken response begins.
