OpenAI has released GPT-Live, a third-generation voice system that eliminates the traditional turn detector by using a full-duplex model capable of listening and speaking simultaneously. This architecture allows for more immediate and natural conversations while maintaining low latency through a new system design optimized for real-time performance.
- The system streams incoming audio into the voice model and outbound speech back to the user, handling deeper reasoning or tool use via frontier models like GPT-5.5 on a separate asynchronous path.
- Media flow is separated from application logic using a dedicated fast path, with inference logic written in Go to improve frame delivery smoothness.
- A seamless handoff mechanism allows for dynamic context compaction and model instance transitions without interrupting the media loop.
- The WebRTC Abridged Roundtrip Protocol (WARP) reduces startup network round trips from six to one.
This foundation powers capabilities in ChatGPT Voice, including computer control and agent coordination in the desktop app, by ensuring that application customizations do not affect the responsiveness of the live media path.