GPT-Live-1: a voice model you can interrupt, now in the API
OpenAI's voice model that listens while it speaks is now in the API. Yelp uses it to answer calls for restaurants and service firms.
What happened?
OpenAI opened its full-duplex voice model GPT-Live-1 to developers through the API on September 10. The model keeps listening while it speaks: callers can interrupt, add a detail mid sentence or change direction without the conversation falling apart. It also handles natural pauses and overlapping speech. API pricing was reported at $0.05 per minute.
What is Yelp doing?
The same week, Yelp announced it had added GPT-Live-1 to two products. Yelp Host is a voice agent that takes phone reservations and orders for restaurants; it launched in October 2025 and had handled more than 1 million calls for restaurants by June 2026, reaching an annual run rate of 2.4 million calls in July. Hatch is a communication platform for home service businesses, which Yelp acquired in January 2026.
According to Yelp, the new model handles background noise, side conversations and interruptions better. It can pick up excitement, impatience or frustration in a caller's tone and switch languages automatically. The company says more calls are resolved without a transfer to a person and that post call transcripts are more accurate, but it gave no figures. Yelp Host costs $249 a month after a free trial of up to 30 days.
How does it work?
GPT-Live-1 handles the voice layer of the call, while details such as reservations, menus and availability come from Yelp's data for each business. The model manages the voice and the flow of the conversation, and the system behind it supplies the right facts. It is the logic of RAG and function calling, applied to the phone.
Why it matters
For small businesses, much lost demand comes from unanswered calls: at peak hours, after hours or when staff are busy with something else. A voice assistant that does not break when interrupted and does not put callers on hold can turn those calls into bookings and appointments. Latency is critical in these products; our latency explainer covers why.
The AI agents QANATONE builds solve the same problem: they answer incoming demand around the clock and route it to the right person.