xAI released Grok Voice 3.0 on October 14, 2026, making its newest speech-to-speech model available in the Voice Agent API at the same price as the version it replaces. Why it matters: voice agents that can hold a real conversation in under a second, at a flat per-minute rate, are getting cheap enough to put on ordinary support lines — not just flagship demos.
October 14, 2026
What happened
Grok Voice 3.0 is now the flagship model in xAI’s Voice Agent API. It costs $0.08 per minute, or $4.80 per hour, unchanged from Grok Voice 2.0. Developers whose agents point at grok-voice-latest were switched over automatically, with no code changes needed, and the previous version has been deprecated in the release notes.
The model handles live, two-way audio and text streaming, which is the setup used for phone agents, in-car assistants and other interactive voice systems.
What the benchmarks say
In independent testing by Artificial Analysis cited in the launch, Grok Voice 3.0 scored 86.9 on the Speech-to-Speech Index, ahead of its predecessor at 81.3. It scored 97% on Full Duplex Bench, which looks at how naturally a model handles pauses, turn-taking and interruptions, and 98.5% on speech reasoning in Big Bench Audio.
On a separate test built from simulated support calls, it reached 71.6% — a big jump from 56.5% for the previous version, though still below one rival configuration on that single test. xAI says its response starts in under a second, and it also reported the highest accuracy, at a rounded 74%, on ServiceNow’s EVA-Bench for enterprise voice agents. Those deployment claims come from the company and have not been independently checked here.
Where it is already running
According to xAI, the model powers voice conversations in Tesla vehicles, where drivers can handle hands-free tasks such as checking a calendar or answering email, and it also runs Grok Bot’s voice mode. The company says Starlink support lines use it for about 20,000 calls a day in multiple languages, with roughly 70% resolved without a person stepping in.
Why it matters
For anyone building or buying a voice agent, the headline is not just the top benchmark score — it is the combination of that score, sub-second replies and no price increase. That puts pressure on rival voice APIs to match on cost per finished task, not just cost per minute. The automatic migration also means many existing Grok Voice customers woke up on a new model today, so real-world call quality over the next few weeks will be the test that matters more than any launch chart.
Frequently asked questions
How much does Grok Voice 3.0 cost?
$0.08 per minute ($4.80 per hour) through the Voice Agent API — the same as Grok Voice 2.0.
Do developers need to change anything to get it?
No, if they use the grok-voice-latest setting. xAI says those agents were migrated automatically.
What is Grok Voice 3.0 best at, on paper?
Live speech-to-speech conversation, especially support calls. It leads the Artificial Analysis Speech-to-Speech Index at 86.9 and scores 97% on Full Duplex Bench for pauses and interruptions.
Sources: xAI (SpaceXSI), Artificial Analysis via xAI, Unite.AI, iClarified