OpenAI brings GPT-Live-1 voice agents to the API

OpenAI released GPT-Live-1 in the API, giving developers access to the company’s full-duplex voice model for apps, phone agents and business workflows.

The practical difference is architecture. Instead of stitching together speech-to-text, a reasoning model and text-to-speech, GPT-Live-1 handles listening and speaking in one model. OpenAI says this improves interruption handling, silence detection, background-noise tolerance and long-session reliability.

The model can also delegate deeper reasoning and tool calls to backend models or third-party systems. That means a voice agent can keep a conversation moving while another model handles tasks such as scheduling, support lookup or complex reasoning.

OpenAI says early tests with Speak cut interruptions by almost 80% compared with prior turn-based systems. The company also says GPT-Live-1 improves its Full Duplex Bench score by 30 percentage points over GPT-Realtime-2.1 and supports telephony deployments for customer support, reservations and other voice workflows.

For builders, the release matters because voice agents are shifting from demos toward production systems. The hard part is no longer just generating natural speech; it is staying responsive while tools, memory and business systems run behind the conversation.