OpenAI launches GPT-Live for live voice conversations in ChatGPT
OpenAI has introduced GPT-Live, a new generation of voice models for ChatGPT. The model makes dialogue with AI feel more like a live conversation: it sounds more natural, holds the context of the conversation better, and is designed for more expressive communication.
A key feature of GPT-Live is its full-duplex architecture, meaning the model can listen to the user and respond at the same time. This should make exchanges less mechanical and help in scenarios such as real-time translation.
For complex requests, GPT-Live will be able to hand off the task to a more powerful model behind the scenes: for example, when internet search, deep analysis, or multi-step work is needed. Rollout has already begun in ChatGPT on iOS, Android, and the web version, with API access promised later.
Why it matters
- —ChatGPT’s voice mode is becoming closer to a natural conversation rather than an exchange of separate commands.
- —The full-duplex approach could improve real-time translation and dialogues where pauses, interruptions, and context matter.
- —Integration with a more powerful model behind the scenes extends the voice interface to complex analytical tasks.
Key facts
- GPT-Live is rolling out in ChatGPT starting today and will become available to all users in the coming days.
- The model will be released on iOS, Android, and the web version of ChatGPT.
- GPT-Live can listen and speak at the same time thanks to its full-duplex architecture.
- For complex requests, the model will be able to delegate work to a more powerful frontier model and return the result in the conversation.
- API access to the new voice models will come later.
The full text is in the original source. Here we provide a brief summary and key facts.
