Grok Voice 3.0 Speech Model
Explore the Grok Voice 3.0 Speech Model: a speech-to-speech engine that resolves support calls on its own, answers in under a second, and cuts cost per task.
Support
Pro AI Tools
Explore elite tools including the Suno AI Music Generator.
Muse Video
Seedance2.5
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
Grok Video Generator
Create Videos from Text or Images with AI
MiniMax H3 video generator
MiniMax H3 AI Video Generator
Seedance 2.0
The Future of AI Video Is Here.
Gemini Omni 1.2 Flash
Meet the Grok Voice 3.0 Speech Model
An end-to-end voice model that resolves more support tickets unaided, replies almost instantly, and already runs at Tesla and SpaceX scale.
- Top-Ranked Speech-to-Speech AccuracyIt finishes more support conversations without a human handoff than any earlier release, and responds quickly enough to sound like someone on the line.
- Frontier Quality at a Smaller BillServing it costs less than rival frontier systems, and the API rate holds steady at the same figure set by the Grok Voice 2.0 launch.
- Proven at Millions of Calls a DayTesla cars on the road and Starlink support desks push millions of live conversations through it every single day.
How to Deploy the Grok Voice 3.0 Speech Model
Four short steps take you from first API call to live voice agents in production.
What the Grok Voice 3.0 Speech Model Does Best
Record-setting scores, human-like turn-taking, and a smaller bill per task for voice agents in production.
First Place on the Speech-to-Speech Index
Independent testing by Artificial Analysis ranks it top with 86.9, a blend of speech reasoning, τ-Voice, blind head-to-head preference, and task success.
97% on Full Duplex Bench
Full Duplex Bench grades how smoothly a model handles turn-taking, pauses, and being cut off — this one scores 97%.
τ-Voice Gains with 40% Less Latency
τ-Voice pushes agents through accented speech, noisy rooms, and interruptions: it reaches 71.6%, level with GPT-Live-1 (Astra) while answering 40% faster.
Cheaper per Completed Task
Task-level figures from Artificial Analysis's Big Bench Audio run put it below both GPT-Live-1 (Astra, Medium) and GPT-Realtime-2.1.
Best EVA-Bench Enterprise Score
On ServiceNow's EVA-Bench for enterprise voice agents it records 74% accuracy, the top figure, plus the strongest combined accuracy-and-experience result.
Automation That Ships in Call Centers
Starlink support desks take 20,000 inbound calls a day through it in several languages; roughly seven in ten close without a human, and a quarter of sales calls convert before hang-up.
Grok Voice 3.0 Speech Model: Common Questions
Quick answers on pricing, response speed, benchmark scores, and real-world deployments for the Grok Voice 3.0 Speech Model.
What exactly is the Grok Voice 3.0 Speech Model?
It is a speech-to-speech system marketed as the strongest of its class, designed to close support conversations that normally need a person, replying in under a second.
How quickly does it respond?
Replies land in less than one second, and τ-Voice comparisons show 40% lower latency than GPT-Live-1 (Astra).
How does it score on independent benchmarks?
It ranks first on the Speech-to-Speech Index at 86.9, with 97% on Full Duplex Bench, 71.6% on τ-Voice, and 74% on ServiceNow's EVA-Bench.
What does it cost to run?
The Voice Agent API bills $0.08 per minute, or $4.80 per hour — unchanged from Grok Voice 2.0 — while per-task cost stays under comparable systems.
Does it cope with interruptions and noisy rooms?
It does. Full Duplex Bench, which measures turn-taking and interruption handling, awards 97%, and τ-Voice covers accented speech plus background noise.
Where is it already running in production?
Millions of conversations a day flow through it in Tesla vehicles and Grok Bot's voice mode, and Starlink call centers route 20,000 calls daily through it in several languages.
Put the Grok Voice 3.0 Speech Model to Work
The Voice Agent API is available now at $0.08 per minute, and any integration already calling grok-voice-latest switches over automatically — nothing to reconfigure.