OpenAI’s GPT-Live-1 puts real AI voice agents on the phone

OpenAI just gave businesses a way to put AI voice agents on the phone that actually sound like they’re listening. The company released GPT-Live-1 in its API on September 11, 2026, and it’s built for full-duplex conversation. That means the model can listen and speak at the same time, the way a real person on a call does, instead of waiting its turn like older voice bots. If you’ve ever hung up on a phone menu because it kept talking over you or missed what you said, this update is aimed directly at that problem.

What actually changed with this voice AI update

Older voice assistants work in turns. You speak, the system processes your words, then it replies. That gap creates the awkward pauses and interruptions you’ve probably experienced with automated phone lines. GPT-Live-1 drops that turn-based structure. It processes audio continuously, so it can respond the instant you finish a sentence, or stay quiet while you’re still talking.

OpenAI reports that early testing showed GPT-Live-1 cut interruptions by close to 80% compared with its previous turn-based voice systems. In plain terms, the AI stops jumping in while a caller is mid-sentence. That single fix matters more than it sounds. Nothing signals “this is a bot” faster than it talking over you, and nothing builds trust faster than a system that waits its turn. You can read OpenAI’s full technical rundown in its announcement post introducing GPT-Live-1.

Where AI voice agents fit into a small business

The obvious use case is the phone line nobody wants to staff at 9pm on a Friday. A restaurant can run reservations through a voice agent that takes a name, party size, and time without a host stepping away from the floor. A clinic can let patients reschedule appointments without a receptionist picking up. A service business can answer “are you open Sunday” calls automatically, freeing your team for the calls that actually need a human.

OpenAI designed the system so the voice layer handles the conversation while a separate backend model handles the thinking. A high-volume booking line can pair the voice layer with a cheaper backend for routine requests, and hand off anything complicated, like a billing dispute or a technical problem, to a stronger reasoning model. That split keeps costs down without making the caller feel like they’ve been dumped into a maze of menus.

This fits a pattern I’ve watched play out across other AI agent launches this year. When we covered OpenAI’s Presence agent platform, the direction was already clear: AI agents are moving from chat windows into the channels your customers already use, starting with the phone.

What it costs and what to check before you build on it

GPT-Live-1 voice is priced at $0.05 per minute, billed per second with no rounding up. For a business fielding a few hundred short calls a month, that’s a real, calculable cost you can compare directly against a part-time receptionist or an answering service. Run the math on your actual call volume before you commit, since a high-volume support line adds up faster than an occasional reservation desk.

If you’re in the EU, don’t skip the compliance side. AI systems that interact directly with customers, including voice agents, fall under the transparency obligations we broke down in our piece on the EU AI Act’s transparency rules. Callers generally need to know they’re talking to an AI system, not a person, so build that disclosure into the call flow from the start rather than bolting it on later.

Frequently Asked Questions

What makes GPT-Live-1 different from older voice bots?

It processes audio in full duplex, meaning it can listen and respond in real time without waiting for you to finish and then processing your speech in a separate step. OpenAI says this cuts interruptions by nearly 80% compared with its earlier turn-based voice systems, so the conversation feels closer to talking with a person.

How much does an AI voice agent built on GPT-Live-1 cost to run?

OpenAI prices the voice layer at $0.05 per minute, billed per second with no rounding. Your total cost depends on call volume and length, so a low-volume reservation line and a high-volume support desk will land very differently on your monthly bill.

Do I need to tell customers they’re talking to an AI voice agent?

If your business operates in the EU, yes. The EU AI Act’s transparency rules generally require you to disclose when a customer is interacting with an AI system rather than a human, and that applies to voice calls just as it applies to chatbots.

Should your business build an AI voice agent now

GPT-Live-1 doesn’t solve every phone problem, but it removes the biggest reason AI voice agents have felt unusable until now: the awkward, interrupting back-and-forth. If your business handles a steady stream of repetitive calls, reservations, appointment changes, hours and pricing questions, this is worth testing on a single phone line before you roll it out everywhere. Start small, measure how many calls actually get resolved without a human, and keep a clear path for callers to reach a person when the AI voice agent hits its limit.