Yelp and Hatch Put GPT-Live-1 Into Voice AI After 1 Million Calls

What happened

OpenAI released GPT-Live-1, a voice model built for full-duplex conversation — the ability to listen and speak at the same time rather than taking rigid turns. The model is now available through the API, and OpenAI is positioning it as a foundation for more natural voice experiences.

The most telling detail for local businesses is not the model itself but who moved quickly to adopt it. Yelp and Hatch have put GPT-Live-1 into their voice AI systems after running roughly one million calls. That is not a pilot. That is a production deployment at a scale where interruptions, background noise, and overlapping speech are the norm rather than the exception.

Full-duplex voice changes the feel of a phone call. Older voice agents waited for silence before responding, which made them sound robotic and caused them to talk over callers or miss quick corrections. A model that can listen while speaking can handle the messy reality of a real conversation: a customer saying "no wait, actually Thursday" mid-sentence, a dog barking in the background, a caller who starts giving their address before the agent finishes asking for it.

The timing matters too. This announcement lands in the same week that Intermedia launched a native AI receptionist, Dialpad partnered to give its phone agents a more human voice, and Zuper added AI agents to field service workflows. Voice AI is no longer a novelty category. It is becoming infrastructure.

Why it matters

For appointment-based local businesses, the phone is still the front door. Dental offices, salons, clinics, repair shops, and studios all live and die by whether someone answers when a customer calls. The gap between a good phone experience and a bad one is often the gap between a booked appointment and a missed one.

What Yelp and Hatch demonstrate is that the technical bar for a natural-sounding call has moved. A million calls is enough volume to expose the failure modes that only show up at scale: accents, poor connections, callers who change their minds halfway through a sentence, and the awkward pauses that make people hang up. If a consumer platform is willing to run that volume through a full-duplex model, the standard for what customers expect from every phone interaction rises with it.

That has a direct commercial consequence. When a caller reaches a voice system that handles overlap and interruption gracefully, they stay on the line. When they reach one that makes them repeat themselves, they hang up and call the next business on the list. The difference is not cosmetic. It is the difference between a filled calendar and an empty one.

There is also a quieter implication. Full-duplex voice makes it harder to tell whether you are talking to a person or a system. That raises the stakes on how a business introduces its phone agent, how it handles a caller who wants a human, and how it captures the details of a booking accurately. Natural conversation is only useful if it ends in a confirmed appointment, not just a pleasant exchange.

What this means for local businesses

The practical lesson is that voice quality is becoming table stakes, not a differentiator. If the underlying models are available to everyone through an API, then the advantage shifts to how a business wires that voice into its actual operations.

A natural-sounding agent that cannot check real availability, book into the right calendar, send a confirmation, and follow up on a no-answer is still just a talking point. The businesses that benefit from this wave will be the ones that treat voice as the start of a workflow, not the end of one. The call has to connect to a calendar, a reminder, and a record of what happened.

It also means the review of any phone system should focus on the unglamorous moments. How does it handle a caller who interrupts? What happens when two people are talking at once? Can it recover when a caller corrects themselves? Those are the moments that decide whether a booking happens, and they are exactly the moments full-duplex models are designed to handle.

For owners who have been putting off a decision on their phone setup, the direction of travel is clear. The consumer platforms are investing in voice because their users expect it. Local businesses will feel that expectation in their own call volumes soon enough.

Full-duplex voice raises the floor on what a phone call should feel like; the businesses that win will be the ones that connect that better conversation to an actual booked appointment.

The bottom line

GPT-Live-1 is a model release, but the Yelp and Hatch deployment is the real story. A million calls through a full-duplex voice system signals that natural phone conversation is moving from demo to default. For appointment-based businesses, the question is no longer whether voice AI can sound human. It is whether the voice on the other end of the line is connected to a calendar, a reminder, and a follow-up that turns a good conversation into a confirmed booking.