Updated
Updated · Engadget · Aug 9
ChatGPT Rolls Out GPT-Live Voice Model, Replacing Turn-Based Mode With Full-Duplex Chat
Updated
Updated · Engadget · Aug 9

ChatGPT Rolls Out GPT-Live Voice Model, Replacing Turn-Based Mode With Full-Duplex Chat

3 articles · Updated · Engadget · Aug 9

Summary

  • GPT-Live is now the default ChatGPT voice model on Android, iOS and the web, replacing the older turn-based Advanced Voice mode with simultaneous listening and speaking.
  • Full-duplex audio lets ChatGPT stop cutting users off after brief pauses and add short acknowledgements such as “mhmm” or “yeah,” making exchanges feel closer to human conversation.
  • GPT-Live-1 is available to Pro, Plus and Go subscribers, while free users get GPT-Live-1 mini; paid users can also switch among Instant, Medium and High intelligence settings.
  • The model can hand off harder reasoning or research tasks to stronger systems such as GPT-5.5 in the background, while keeping the voice conversation flowing and showing transcripts or visual widgets when useful.
  • Screen and video sharing are not yet supported, though users can still revert to the older voice model in settings as OpenAI pushes voice interaction further beyond the 2023 speech-to-text pipeline.

Insights

Could GPT-Live's hidden separation from your main text chat secretly ruin your professional AI workflows?
Why does ChatGPT's new GPT-Live mode force human-like acknowledgments that users cannot disable?
As OpenAI introduces GPT-Live, why are the actual latency figures for its background GPT-5.5 model kept secret?