OpenAI on Wednesday launched GPT-Live, a new family of voice models built to make spoken exchanges with artificial intelligence feel more natural. The company said these models now serve as the engine behind ChatGPT Voice.
GPT-Live is based on a full-duplex design that permits simultaneous listening and speaking. That architecture lets the system provide brief verbal acknowledgements like "mhmm" or "yeah," take part in quick back-and-forths, or remain quiet when appropriate, according to OpenAI.
The company also described a division of labor between GPT-Live and higher-capability models. Tasks that require web searches, more intensive reasoning, or multifaceted work can be delegated to frontier models running in the background. At launch, those background tasks will be handled by GPT-5.5, with OpenAI planning to switch to newer frontier models as they become available.
OpenAI is distributing two initial variants of GPT-Live: GPT-Live-1 and GPT-Live-1 mini. Both began rolling out to ChatGPT users worldwide on Wednesday, and the company said it intends to offer API access to the models in the near term.
In contrast to earlier voice systems that either chained models in sequence or required users to finish speaking before the system responded, GPT-Live continuously processes incoming audio while generating output. OpenAI noted that the model makes interaction decisions multiple times per second, enabling a more fluid approach to turn-taking.
In direct evaluations, GPT-Live-1 and GPT-Live-1 mini were chosen over the previous Advanced Voice Mode in conversations that lasted five to ten minutes. The tests assessed overall preference plus specific dynamics such as turn-taking, interruptions, conversational flow, and how natural the interactions felt.
OpenAI reported that more than 150 million people use ChatGPT Voice and Dictation features on a weekly basis. The updated ChatGPT Voice experience that incorporates GPT-Live includes what the company described as more natural conversation, smarter responses through access to frontier models, enhanced listening performance, and the ability to provide visual responses.
Users are given choices for reasoning intensity: Instant for rapid replies, or Medium and High settings for deeper deliberation. While speaking, users can receive visual cards covering topics such as weather, stocks, and sports. The voice feature continues to support search, memory, images, and file uploads.
Regarding availability, GPT-Live-1 will be the default voice model for Go, Plus, and Pro subscribers, and GPT-Live-1 mini will be the default for Free users. The models are accessible on iOS, Android, and ChatGPT.com.
OpenAI said GPT-Live will not support voice with video or screen sharing in ChatGPT at launch but that work is underway to add those capabilities. Users who require voice paired with video or screen sharing can still use earlier versions of ChatGPT Voice where those features are present.