OpenAI describes GPT-Live, its third-generation voice system, which removes the traditional turn-detector component and instead streams audio continuously through a full-duplex model capable of listening and speaking simultaneously. Audio moves through a dedicated fast path while deeper reasoning or tool use is delegated asynchronously to frontier models like GPT-5.5 without interrupting conversation flow. The team also reduced voice-session startup from six network round trips to one to cut latency.
