GPT‑Live
Points and comments are a snapshot, not live.
OpenAI launches GPT-Live, a full-duplex voice AI for more natural, real-time conversation.
OpenAI introduced GPT-Live, a voice model built on a full-duplex architecture that listens and speaks simultaneously, enabling natural back-and-forth without waiting for turns. It delegates deeper tasks (search, reasoning) to GPT-5.5 in the background, maintaining conversation flow. Two versions, GPT-Live-1 and GPT-Live-1 mini, roll out today to ChatGPT users globally, with API access planned. The model outperforms Advanced Voice Mode in human evaluations measuring pleasantness and turn-taking. Safety measures include real-time safeguards, emotional reliance monitoring, and predefined voices to prevent impersonation.
What commenters are saying
Commenters are excited about the potential for language learning and real-time translation, though some note the French translation was barely understandable and heavily accented. A recurring complaint is that the previous Advanced Voice Mode interrupted too easily, but some worry the new version may interrupt users instead. Early testers report hour-long productive conversations, but note quirks like laughing at unintended jokes. Several ask about open-source alternatives, with one developer working on an open-weights full-duplex model at duplexio.ai. No pricing announced yet.