OpenAI has introduced GPT-Live, a new generation of voice models that now powers ChatGPT Voice. The feature is now rolling out globally to iOS, Android, and ChatGPT.com users.
OpenAI has introduced GPT-Live, a new generation of voice models that now powers ChatGPT Voice. The feature is now rolling out globally to iOS, Android, and ChatGPT.com users.
The main difference of GPT-Live is its full-duplex architecture: the model can listen and speak simultaneously, rather than one at a time, as it did before. This makes the conversation more natural: during a dialogue, GPT-Live can show that it is listening attentively with short remarks like «uh-huh» or «yes», maintain a rapid exchange of remarks, or simply remain silent when the interlocutor needs time to think.
It’s also OpenAI’s smartest voice model to date, the company says. If a query requires web search, deeper reasoning, or more complex work, GPT-Live seamlessly delegates the task to the flagship text model behind the scenes—GPT-5.5 at launch—and returns the result to the conversation when it’s ready. While processing is in progress, the voice model continues to support the conversation so as not to interrupt its flow.
OpenAI announced the launch of long-term post-release monitoring focused on users' emotional attachment — a continuation of the company’s previous research on the affective use of chatbots and emotional well-being.
Separately, the company emphasized that GPT-Live was created specifically for conversation, and not for imitating the voice of a specific person: the model uses a set of predefined voices in ChatGPT, and is protected against reproducing the voice of a real person.
GPT-Live-1 becomes the default voice model for Go, Plus, and Pro plans, and GPT-Live-1 mini for free users. The model is optimized for the most popular languages in ChatGPT, although the company warns that for some languages, there may be non-native accents or gaps in fluency. GPT-Live does not currently support voice chat with video or screen sharing — OpenAI promises to add this feature in the near future.
