Breaking

xAI's Grok Voice APIs Go Live on Vercel AI Gateway

June 29, 2026 at 14:24 EDT

Foundation Models
AI Agents
Software Dev & Coding

On June 29, 2026, xAI announced that its Grok-based real-time voice APIs—covering voice agents, text-to-speech and speech-to-text—are now available through Vercel's unified AI Gateway. The move lets developers add xAI's voice features to their apps while handling routing, monitoring and spend management through a single API key and dashboard.

xAI · Vercel AI Gateway

Grok Realtime Voice Lands on the Vercel AI Gateway

Three audio models — realtime voice, TTS and STT — are now callable through a single endpoint and one API key, slotting into Vercel AI SDK 7 with a browser useRealtime hook for low-latency voice over WebSocket.

780ms

Time to First Audio (p50)

67.3%

τ-voice Bench score (leads)

97.1%

Big Bench Audio

Models: voice · TTS · STT

One model vs. the classic chain

grok-voice-think-fast-1.0 fuses voice I/O and reasoning into a single model — collapsing the traditional three-step pipeline.

STT

LLM

TTS

Classic chain · 3 stages

Voice + Reasoning

Single model · 1 stage

Proven in production · Starlink support

xAI's voice already powers Starlink's phone support and sales lines.

70%

of inquiries resolved automatically

20%

sales conversion rate

Pricing & specs

Realtime voice

$0.05/min

connected-time billing

Speech-to-text

$0.10/hr

batch · streaming $0.20/hr

Text-to-speech

$4.20/1M chars

5 voices · 20+ languages

Capacity

25+ langs · 28

concurrent tool calls

Why it matters

Polished for complex workflows — fast interruption handling and background reasoning enable natural conversation, callable with minimal extra code.

Still to prove

The Vercel integration is new and the realtime feature retains experimental elements; long-term stability and diverse-accent handling await broad testing.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.