The company said GPT-Live is now rolling out globally across iOS, Android and ChatGPT.com, where it will power the new ChatGPT Voice experience.
Unlike previous versions of ChatGPT Voice, which waited for users to finish speaking before responding, GPT-Live uses a full-duplex architecture that enables continuous interaction. This allows the AI to acknowledge users with brief responses such as “mhmm” or “yeah”, pause when users are thinking, or interrupt naturally when appropriate.
OpenAI said the new architecture also makes conversations feel less rigid and more like speaking with another person.
For more demanding tasks such as web searches, deeper reasoning or complex problem-solving, GPT-Live delegates the work to OpenAI’s latest frontier models while continuing the conversation. At launch, those background tasks will be handled by GPT-5.5, allowing users to keep talking instead of waiting silently for a response.
The company said GPT-Live is its smartest voice model to date and will automatically benefit from future frontier model upgrades.
OpenAI is initially launching two versions of the model—GPT-Live-1 and GPT-Live-1 mini. GPT-Live-1 will become the default voice model for Go, Plus and Pro subscribers, while GPT-Live-1 mini will power ChatGPT Voice for free users.
The upgraded ChatGPT Voice experience also introduces richer visual responses. During conversations, users can now see information cards for topics such as weather, stocks and sports while continuing to interact through voice.
OpenAI said GPT-Live improves handling of background noise and is less likely to interrupt users during natural pauses in conversation. Users can also ask the assistant to remain silent and simply listen when needed.
According to the company, internal evaluations showed GPT-Live was preferred over the existing Advanced Voice Mode across measures including conversational flow, turn-taking, interruptions and overall naturalness. It also demonstrated stronger performance on scientific reasoning, web search and multi-step customer support tasks.
The company said GPT-Live has been designed with dedicated voice-specific safety measures, including safeguards for conversations involving self-harm, emotional reliance, violence and inappropriate content. It also includes protections against voice impersonation by restricting responses to OpenAI’s predefined voices.
At launch, GPT-Live will not support voice conversations with video or screen sharing, though OpenAI said those capabilities are being developed and existing voice modes will continue to support them in the meantime.