On July 8, 2026, OpenAI launched GPT-Live, a new generation of voice models built on a full-duplex architecture that can listen and speak at the same time. Instead of the rigid turn-taking of earlier voice systems, GPT-Live can backchannel with phrases like “mhmm” while the user talks, engage in quick back-and-forth, or stay quiet when the user pauses to think.
GPT-Live also changes how voice connects to frontier intelligence. For questions that need web search, deeper reasoning, or complex work, the voice model delegates to OpenAI’s latest frontier model behind the scenes and folds the result back into the conversation, keeping the dialogue flowing while the heavier model works. At launch GPT-Live uses GPT-5.5 in the background, with the backing model to be updated as new frontier models ship.
OpenAI framed the release as the third architectural generation of its voice stack: the original ChatGPT Voice chained speech-to-text, an LLM, and text-to-speech; Advanced Voice Mode processed audio in a single model but still operated in discrete turns detected by silence; GPT-Live removes the turn boundary entirely. Two versions, GPT-Live-1 and GPT-Live-1 mini, began rolling out to ChatGPT users globally on launch day, with API access planned.