'GPT-Live' Arrives, Bringing Interruptible Full-Duplex Voice Conversations to ChatGPT
OpenAI has unveiled “GPT-Live,” a new-generation voice model. It adopts a full-duplex architecture that lets it listen and speak at the same time, delivering much more natural voice conversations. It launched worldwide immediately as ChatGPT’s voice feature.
Details
- Natural conversation flow: Can keep a conversation going while interjecting backchannel cues like “mm-hmm” or “yeah,” and handles interruptions naturally. It also recognizes when the user needs a moment to think
- Full-duplex processing: Makes decisions about speaking, listening, waiting, or interrupting multiple times per second
- Task delegation: Delegates web search and complex reasoning to separate backend models while maintaining conversational continuity
- Multimodal support: Displays visual cards for weather, stock prices, sports scores, and more during a voice conversation
- Noise handling: Improved background noise filtering reduces false interruption detection
- Limitations: Currently doesn’t support video or screen sharing. The previous voice mode remains available as well
Availability
| User tier | Model |
|---|---|
| Paid plans | GPT-Live-1 (with Instant/Medium/High reasoning levels) |
| Free plan | GPT-Live-1 mini |
How to try it
- Available on iOS, Android, and ChatGPT.com
- Targets the 150 million users who use voice features weekly
- An API version is planned for the future; sign-ups for developers and enterprises have also opened