OpenAI's GPT-Live gives ChatGPT full-duplex voice that talks back in real time

OpenAI has launched GPT-Live, a new class of voice models built on a full-duplex architecture that lets ChatGPT listen and speak at the same time rather than waiting for a turn to end. The practical effect is conversation that behaves less like a walkie-talkie and more like a phone call: the model can drop in backchannel cues like ‘mhmm’ or ‘yeah,’ handle rapid interruptions, and stay silent when the user needs a moment to think.

The more interesting design choice is what happens under the hood. When a request needs web search, heavier reasoning, or other slow work, GPT-Live hands off to a frontier model in the background and keeps the spoken conversation going, folding the result back in once it’s ready. That decoupling of the low-latency voice loop from the expensive reasoning path is how OpenAI keeps the interaction feeling continuous instead of stalling on every complex query.

Rollout is global and immediate across web and mobile, in two tiers: GPT-Live-1 becomes the default voice model for Go, Plus, and Pro subscribers, while GPT-Live-1 mini serves Free users. The move lands alongside OpenAI’s broader GPT-5.6 expansion and signals that always-on, interruptible voice is becoming the default interface rather than a novelty mode.