OpenAI launches GPT-Live-1, a full-duplex voice model, in the API

OpenAI put GPT-Live-1, a full-duplex voice model that can listen and speak at the same time, into the developer API at $0.05 a minute for the voice layer.

A voice model that can listen and talk at once, like a phone call, is now in the developer API at $0.05 a minute. The interruption-cut and bench gains stay OpenAI’s.

OpenAI published “Build more natural voice experiences with GPT-Live-1 in the API.” The same day’s API changelog listed gpt-live-1 as generally available on the live-session route developers call. Endpoint IDs are in Sources.

OpenAI says GPT-Live-1 is a full-duplex voice model, meaning it can listen and speak at the same time, the way a phone call works. Interruptions are handled in one model, not a chained speech-to-text / large language model (LLM) / text-to-speech stack that waits for each step to finish.

Price on the launch post and the model card: $0.05 per minute for the front-end voice layer, billed per second — five cents for each sixty seconds of that voice path. Backend model and tool usage is charged separately.

Developer docs: the voice model can delegate reasoning and tool use to a backend agent — a managed Responses model, or the app’s own agent via client delegation. Interrupting speech does not automatically cancel that backend work.

This is a developer product announcement, not a consumer ChatGPT update. The Speak “almost 80% fewer interruptions” line and the Full Duplex Bench gains are OpenAI’s customer and lab numbers.