OpenAI Ships GPT-Live-1 Full-Duplex Voice Model in the API

GPT-Live-1, which OpenAI made available in its API on September 10, is a voice model built to listen and speak at the same time. OpenAI calls this full duplex: instead of taking strict turns, the model is meant to keep hearing the caller while it talks, so a person can jump in mid-sentence, and the launch video says it is built to handle background noise and interruptions smoothly. The post pitches it as bringing ChatGPT's natural back-and-forth to developers' own apps.

The key design point is that GPT-Live-1 is only the conversational front end. In the launch video, OpenAI describes it as pairing with a separate backend model for reasoning and tool calls, which is how the demo has it hand actions to a robot and a display while the conversation keeps going, and the post says it works with the models and harness the developer chooses. That split also shapes the price OpenAI quotes in the video: five cents per minute for the GPT-Live-1 front end, with backend inference and tool services charged separately. The per-minute figure is therefore not the full cost of a voice agent, which depends on the backend paired with it.

The noise and interruption handling shown here comes from OpenAI's own launch demo, not independent testing. This explanation relies on the post and an automatic speech transcript of the video; the visual parts of the demo were not reviewed. Separate coverage of the same release draws on OpenAI's written API announcement.