OpenAI has started rolling out GPT-Live, its third-generation voice model and architecture, to ChatGPT users worldwide. Free users get GPT-Live-1 mini, while paid users default to GPT-Live-1. The system separates complex reasoning from real-time conversation, letting it listen while the user speaks — no more walkie-talkie-style turn-taking.
https://twitter.com/OpenAI/status/2074907025537224840
GPT-Live uses a full-duplex architecture, supporting two-way listening and speaking. During a conversation, it will use vocal cues like "uh-huh" and "right" to signal it’s paying attention, allows for rapid back-and-forth, or stays quiet when the user needs a moment to think.
When faced with a question that requires search or deep reasoning, GPT-Live delegates the task to GPT-5.5 in the background while continuing to talk to the user. It can also present visual cards for things like weather and stock prices to aid understanding.