xAI launched Grok Voice Think Fast 1.0, a new real-time voice-agent model exposed through the xAI API.
What it is for
xAI targets the model at messy real-world voice workflows:
- customer support
- phone sales
- appointment booking
- restaurant reservations
- multi-step enterprise support flows
- high-volume tool calling
The model is designed to collect and confirm structured data such as addresses, phone numbers, account numbers, and email addresses from spoken conversation.
Benchmark and language claims
xAI says Grok Voice Think Fast 1.0 ranks first on tau-voice Bench, a benchmark for full-duplex voice agents under noisy, accented, interrupted, realistic conversation conditions.
The model supports 25+ languages.
Starlink deployment
xAI says the model powers Starlink’s phone sales and customer support experience.
Reported operational metrics:
- 20% conversion rate on sales inquiries.
- 70% autonomous resolution rate for support inquiries.
- 28 tools used by the single voice agent.
Why it matters
Grok has been mostly judged as a chat model and X-integrated assistant. This release gives xAI a stronger enterprise voice story and makes Grok more directly comparable with OpenAI Realtime, ElevenLabs Conversational AI, and enterprise contact-center agents.
The timing is awkward: the launch landed during a week of Grok congestion reports. That makes reliability the main question for buyers.
Related
Sources
Primary and corroborating references used for this news item.