OpenAI made GPT-Live-1 generally available in its API on September 10, 2026. It supports full-duplex voice conversations: the model can listen while speaking instead of requiring callers to wait for a fixed turn to finish.
GPT-Live handles the audio conversation while a backend model or agent performs reasoning and uses tools. Developers can delegate those tasks to an OpenAI model through the Responses API or connect their own backend. The voice and backend models are chosen separately.
The September 10 changelog lists voice sessions at $0.05 per minute, billed per second. Backend model and tool usage is charged separately.
OpenAI's integration guide warns that an existing Realtime integration is not automatically compatible with GPT-Live. Supported package versions, interruptions and backend delegation need to be checked for the chosen voice framework or phone provider.
Interrupting the assistant does not necessarily cancel work already underway. The application controls whether the backend finishes or cancels the task, according to OpenAI's GPT-Live documentation.