Overview
- OpenAI replaced turn‑detector driven voice turns with GPT‑Live, a full‑duplex system that listens and speaks at the same time and decides multiple times per second whether to listen, speak, pause, or allow interruptions.
- Engineers split the service into a low‑latency fast path for instant audio streaming and an asynchronous background path for heavier work so the voice layer stays smooth while deeper reasoning completes.
- The team rewrote transport code from Python asyncio to Go to deliver audio frames more consistently at scale for the platform's large voice user base.
- OpenAI published a new open specification called WARP and an Instant Connect feature that compress WebRTC startup down to a single packet to cut connection latency and speed session launch.
- GPT‑Live is rolling out across ChatGPT Voice on iOS, Android, and the web with GPT‑Live‑1 for paid plans and GPT‑Live‑1 mini for free users, and API access is planned later which will shape enterprise and developer use.