OpenAI launches GPT-Live-1 in the API for full-duplex voice agents

GPT-Live-1 is now available through the API as a single voice model that can listen and speak at the same time. It can handle interruptions more naturally, delegate deeper reasoning and tool calls to backend models, support phone-based voice agents, and give developers more control over tone, pace, and conversational style.

Voice-agent stacks usually chain speech-to-text, a reasoning model, and text-to-speech, which adds latency and coordination complexity. GPT-Live-1 moves more of the real-time conversation into one model, which can make production voice assistants feel more natural and simplify the architecture developers need to maintain.

Yes — Worth testing for call-center, booking, support, and other real-time voice workflows where interruption handling and latency matter.