OpenAI makes ChatGPT Voice more natural with GPT-Live

• This photo was AI-generated for publication use only.

OpenAI has introduced GPT-Live, a new generation of voice models designed to make conversations with ChatGPT feel more natural, faster, and closer to real human interaction.

The new model now powers ChatGPT Voice and allows the AI to listen and speak at the same time through what OpenAI calls a full-duplex architecture. This means ChatGPT can follow a conversation while a user is still speaking, respond with short acknowledgments such as “mhmm” or “yeah,” pause when needed, or stay quiet when the user needs more time to think.

OpenAI said GPT-Live is also its smartest voice model so far. For more complex questions that require web search, deeper reasoning, or other advanced tasks, GPT-Live can delegate the work to GPT-5.5 in the background while keeping the voice conversation going.

The company said this allows ChatGPT Voice to provide smarter answers without making conversations feel slow or rigid.

GPT-Live will roll out in two versions: GPT-Live-1 and GPT-Live-1 mini. GPT-Live-1 will become the default model for ChatGPT Voice for Go, Plus, and Pro users, while GPT-Live-1 mini will be the default for Free users.

The rollout is available globally across iOS, Android, and ChatGPT.com. OpenAI also plans to bring GPT-Live to the API soon for developers and enterprises.

With the update, ChatGPT Voice can now handle more natural back-and-forth conversations, better understand pauses, and focus more clearly on the user’s voice even when there is background noise.

OpenAI said ChatGPT Voice can also show visual cards while users are talking, including information on weather, stocks, sports, and other topics. Voice will continue to support search, memory, images, and file uploads.

The company said more than 150 million people use ChatGPT Voice and Dictation every week for hands-free help, language practice, storytelling, commuting, and everyday conversations.

OpenAI also emphasized safety features built for real-time voice use. These include safeguards that can steer responses when potentially unsafe output is detected, provide safety messages or resources, or end a voice conversation in higher-risk cases.

The company said GPT-Live was designed for conversation and not for voice impersonation. It uses predefined voices in ChatGPT and includes safeguards to prevent the system from imitating real people.

At launch, GPT-Live will not yet support voice with video or screen sharing in ChatGPT, though OpenAI said it is working to add those capabilities soon. The company also noted that some languages may still have non-native accents or fluency gaps as it continues improving the experience.