OpenAI launches full-duplex GPT-Live voice models as default across ChatGPT tiers

OpenAI launched GPT-Live on July 8, a new generation of voice models built on a full-duplex architecture that lets the system listen and speak at the same time. The two models, GPT-Live-1 and GPT-Live-1 mini, let users interrupt mid-sentence without the system talking over them or cutting off.

The move is OpenAI’s clearest bet yet that voice, not text, becomes the main way people use ChatGPT. Making the new models the default across every subscription tier puts the wager in front of hundreds of millions of users at once.

GPT-Live-1 becomes the default voice model for ChatGPT Go, Plus and Pro subscribers, while GPT-Live-1 mini becomes the default for Free users, OpenAI said, with the rollout starting globally the same day.

The design splits conversation from heavy reasoning. When a query needs web search or deeper analysis, GPT-Live routes it in the background to OpenAI’s GPT-5.5 frontier model and returns the result to the ongoing conversation, keeping the spoken exchange responsive. OpenAI said it will keep updating that underlying frontier model over time, so the voice layer can improve without a new voice release.

Full-duplex voice has been a hard problem for the field: earlier assistants relied on turn-taking that produced stilted, walkie-talkie exchanges. Whether GPT-Live feels natural in noisy, real-world use — and how it handles latency when it hands off to GPT-5.5 — will decide if voice-first sticks or stays a demo. OpenAI has not published independent latency or accuracy figures for the new models.