Full-duplex conversation
GPT-live-1 can listen while it speaks, which makes interruptions and quick back-and-forth exchanges feel much closer to a natural call.
GPT-live-1 · OFFICIAL API
Natural full-duplex AI voice in your browser. Speak, interrupt, change direction and keep the conversation moving without a record-transcribe-playback loop.
READY
GPT-live-1Full-duplex voice over WebRTC1 credit reserves 1 minute. End early and unused reserved whole minutes are returned automatically.
WHAT IS GPT-LIVE-1?
GPT-live-1 is OpenAI’s real-time voice model for natural spoken conversation. This browser workspace connects your microphone to the official GPT-live-1 API over WebRTC, so the model can listen and speak with low latency and handle interruptions more naturally than a turn-by-turn transcription pipeline. One credit reserves one minute of live conversation, and unused reserved minutes are returned when you end early.
GPT-live-1 can listen while it speaks, which makes interruptions and quick back-and-forth exchanges feel much closer to a natural call.
The workspace uses WebRTC rather than uploading recorded audio clips. Your microphone connects to an official OpenAI Live session after the server authorizes it.
Choose a maximum session length before connecting. Those minutes are reserved from your balance, then unused whole minutes are returned when you finish early.
HOW IT WORKS
Instead of recording a clip, waiting for transcription, generating text and then synthesizing speech, GPT-live-1 keeps one live audio session open so speaking and listening can overlap.
Choose a maximum length. One credit reserves one minute, so you always know the most the session can consume before the microphone opens.
Allow microphone access and speak normally. GPT-live-1 responds with live audio over WebRTC and can handle interruptions without waiting for a separate upload/transcribe step.
End the call whenever you are done. Billing rounds used time up to the next minute and automatically returns the unused part of your reservation.
LIVE SESSION DETAILS
A live microphone/audio session rather than file upload and polling.
Minutes are reserved before connecting; unused reserved whole minutes are returned after an early end.
The Live session is created against OpenAI’s official gpt-live-1 model.
FAQ
GPT-live-1 is OpenAI’s real-time voice model for natural full-duplex spoken conversations. It is designed to listen and speak with low latency, including while a conversation is being interrupted.
No. GPT-live-1.pro is an independent web application that connects to the official OpenAI GPT-live-1 API. It is not operated by or endorsed by OpenAI.
One credit reserves one minute of a GPT-live-1 session. There is no free trial. Prepaid minute packs are shown on the pricing page.
A live WebRTC session can continue until it is ended, so reserving a maximum length creates a clear spending ceiling. If you stop early, unused reserved whole minutes are returned automatically.
Yes. Natural interruption handling is one of the main differences between GPT-live-1 and a traditional record, transcribe, answer, and text-to-speech pipeline.
No. The API key stays on the server. The browser sends its WebRTC offer to this site, and the server creates the authorized GPT-live-1 session and returns only the WebRTC answer.
This workspace does not upload a recording to its own media library. Live audio is transported through the WebRTC session to the model provider. Avoid discussing confidential information you are not permitted to process.
The session is ended at its reserved limit so it cannot continue consuming paid live time beyond the amount you selected.