OpenAI Launches GPT-Live Voice System That Listens and Speaks Simultaneously for Near-Instant Conversation
Summary
OpenAI launches GPT-Live, a revolutionary full-duplex voice system that lets AI listen and speak simultaneously, delivering near-instant, natural conversation by eliminating traditional turn detection and cutting session startup from six network round trips to just one.
Key Points
- OpenAI's GPT-Live, a third-generation full-duplex voice system, eliminates the traditional turn detector from the audio path, allowing the model to listen and speak simultaneously for more natural, low-latency conversation.
- The system uses a dedicated media fast path built in Go with WebRTC transport, asynchronous delegation to frontier models like GPT-5.5, and seamless stateful inference handoffs to keep audio flowing continuously without interruption.
- A new protocol called WARP reduces WebRTC session startup from six network round trips to one, while a shadow testing strategy using real production traffic revealed and resolved critical capacity, geography, and long-session reliability issues before public launch.