Grok Voice 3.0
Talk to Grok Voice 3.0 in real time: natural, full-duplex speech that answers calls in under a second and costs just $0.08 per minute.
Support
Pro AI Tools
Explore elite tools including the Suno AI Music Generator.
MiniMax H3
MiniMax H3 AI Video Generator
Seedance 2.5
The Future of AI Video Is Here.
Seedance 2.0
The Future of AI Video Is Here.
Veo3.1
Create Stunning Videos with Veo3.1
FLUX 3 Video Generator
Kling 3.0
Next-Gen AI Video Generator
Grok Video Generator
Create Videos from Text or Images with AI
MiniMax H3 video generator
MiniMax H3 AI Video Generator
Inside the Grok Voice 3.0 Speech Engine
An always-on voice engine built for live calls: Grok Voice 3.0 listens, reasons and answers in a single pass, so conversations feel like talking to a person rather than a script.
- Benchmark-Leading Call AccuracyIndependent testing places Grok Voice 3.0 at the top of Artificial Analysis's Speech-to-Speech Index with an 86.9 score, ahead of every rival model evaluated.
- Conversations That Never StallReplies land in less than a second, and a 97% Full Duplex Bench result shows the model handles pauses, barge-ins and hand-offs the way people do.
- Battle-Tested at Global ScaleEvery day it carries millions of exchanges across Tesla and SpaceX fleets, among them 20,000 Starlink support calls — around 70% closed with no agent on the line.
Putting Grok Voice 3.0 to Work
From the first API call to round-the-clock multilingual support — the shortest path from sign-up to a live voice agent.
Grok Voice 3.0 Benchmarks That Matter
Accuracy, speed, resilience and spend per task — the four numbers that decide whether a voice agent survives contact with real customers.
Best-in-Class Call Resolution
Grok Voice 3.0 holds the highest score ever recorded on Artificial Analysis's Speech-to-Speech Index (86.9) and closes more support calls end-to-end than any prior release.
Human-Like Turn Taking
Full Duplex Bench awards it 97% for the way it waits, yields and recovers when a caller cuts in mid-sentence.
Steady Under Accents and Noise
On τ-Voice, which throws accented speech, noisy backgrounds and interruptions at the agent, Grok Voice 3.0 scores 71.6%.
Smart and Quick at the Same Time
It matches GPT-Live-1 (Astra) on task quality in τ-Voice yet returns answers 40% sooner, closing the usual trade-off between reasoning depth and speed.
Cheaper per Completed Task
Artificial Analysis's Big Bench Audio run shows a lower bill for every task finished than either GPT-Live-1 (Astra, Medium) or GPT-Realtime-2.1.
Trusted on Enterprise Benchmarks
ServiceNow's EVA-Bench gives it a 74% accuracy mark — the strongest of any model tested — alongside the best combined accuracy and customer-experience figure.
Grok Voice 3.0: Common Questions
Quick answers on how Grok Voice 3.0 performs, what it costs, and what happens to agents you have already shipped.
What exactly is Grok Voice 3.0?
A speech-to-speech model that hears and speaks in a single pass. It is live in the API now, and the rate is identical to the version it replaces.
How well does it score on benchmarks?
Artificial Analysis places it first on its Speech-to-Speech Index at 86.9, and Full Duplex Bench records 97% for conversational behaviour.
How quickly does it respond mid-call?
Round-trip delay stays below one second, and on τ-Voice it comes back 40% faster than GPT-Live-1 (Astra).
What does Grok Voice 3.0 charge?
Billing is unchanged at eight cents a minute, or $4.80 an hour, on the Voice Agent API — the same figure as the previous generation.
Does it cope with live customer support?
It already does. Starlink routes 20,000 calls a day through it in several languages, and roughly 70% of those finish without a person taking over.
Must I migrate agents already in production?
Not if they point at grok-voice-latest — the handover completed in the background, and nothing in your integration needs touching.
Ship Your Voice Agent with Grok Voice 3.0
Spin up a live call in minutes and hear the difference yourself — answers in under a second, enterprise-grade resolution, and eight cents a minute.