OpenAI's GPT-Live Delegates Reasoning While It Talks

GPT-Live processes speech and generates output at once, deciding many times a second whether to talk, listen, or hand off, per OpenAI.

That matters for anyone supervising agents that keep working after you stop typing. OpenAI is pitching voice, not a dashboard, as the interface for checking in on that work.

That's a break from turn-based assistants, which go silent the moment they start responding. When a question needs a web search or harder reasoning, GPT-Live hands the work to OpenAI's frontier model in the background. It keeps talking the whole time, folding the answer in once it's ready.

The model is the default voice for Go, Plus, and Pro accounts. A lighter mini version covers free users, and OpenAI doesn't say whether it gets the same background delegation, or a scaled-down version of it.

OpenAI is positioning voice as the interface for supervising long-running agentic work. It's the layer you talk to while a task runs in the background, instead of a chat window you keep refreshing.