OpenAI GPT-Live-1 is a full-duplex voice model for developers building real-time voice agents that listen and speak at the same time. Released through the OpenAI API on September 10, 2026, it replaces the older turn-based Advanced Voice Mode with a single model that reasons over incoming and outgoing audio together.
Core Features
- Full-duplex audio: the model handles pauses, interruptions, and backchannels in one layer instead of chaining separate speech-to-text, language-model, and text-to-speech steps.
- Backend delegation: GPT-Live-1 keeps the conversation flowing while handing deeper reasoning and tool calls to a model such as GPT-6 Astra, Codex, or a third-party backend.
- Production controls: native ASR transcripts, explicit turn detection, keyword biasing, and better handling of silence and background noise.
- Connectivity: stream over WebRTC for browsers, WebSockets for server-side audio, or Telephony and SIP for phone agents.
- Twelve new real-time voices (Quartz, Ripple, Vesper, Willow, and others) with accents, dialects, and configurable tone and pace through the system prompt.
Use Cases
- Customer-support lines that need to handle mid-sentence interruptions without dropping context.
- Phone-based reservations and order updates where the agent must call APIs in the background.
- Language tutoring and accessibility tools that benefit from low-latency, natural back-and-forth speech.
Pricing
The front-end voice layer costs $0.05 per minute, billed per second; backend model and tool usage is charged separately. The free tier is not supported, and concurrent-session limits scale with your usage tier. Explore more options in our AI audio tools directory.
Our Take
Best for teams shipping production phone and voice assistants that must survive interruptions. The trade-off is that GPT-Live-1 has no image or video input and no structured outputs, so you still need a backend model for anything beyond speech.
