OpenAI Launches GPT-Live - a Full-Duplex Voice Model That Talks While It Listens
GPT-Live replaces Advanced Voice Mode with a full-duplex architecture that can listen and speak simultaneously and hand off hard questions to GPT-5.5 in the background.
OpenAI launched GPT-Live today, a new generation of voice models that replaces ChatGPT Advanced Voice Mode with a full-duplex architecture - meaning it can listen and speak at the same time. GPT-Live rolls out now to ChatGPT users globally on iOS, Android, and the web. GPT-Live-1 becomes the default for Go, Plus, and Pro subscribers; GPT-Live-1 mini takes over for Free users.
Full-Duplex and Background Delegation
Previous ChatGPT voice operated in turns: the model waited for silence, then responded. GPT-Live drops that constraint. Built on a full-duplex architecture, it processes input continuously while generating output - making interaction decisions many times per second on whether to speak, pause, or wait. Back-channel responses like "mhmm" and "got it" keep conversations feeling live rather than transactional.
GPT-Live also separates fast conversation from heavy reasoning. When a question requires web search or deeper thinking, GPT-Live delegates to GPT-5.5 running in the background and keeps talking while the model works. At launch it uses GPT-5.5; OpenAI says it will update the background model as new ones ship. This is a meaningful architectural shift - GPT-5.6 already ships in three tiers, and the delegation design means GPT-Live can inherit improvements from those releases without a separate voice model update.
What's New in ChatGPT Voice
All nine ChatGPT voices have been remastered for GPT-Live. Four reasoning tiers are available: Instant and mini for fast responses, Medium and High for tasks where GPT-5.5 Thinking handles the background work with more compute. Visual cards now appear during voice conversations for weather, stocks, sports scores, and maps - answers you'd otherwise have to ask to be read aloud. Background noise handling is also improved; the model waits when you pause rather than jumping in on silence.
ChatGPT reached 1 billion monthly users earlier this year, with 150 million people using Voice weekly. GPT-Live is OpenAI's first full architectural replacement for the voice layer since Advanced Voice Mode shipped in 2024. The API is not available at launch - developers can sign up to be notified. Video and screen-sharing support is also absent at launch; legacy voice modes remain available for those features.