OpenAI GPT-Live Brings Continuous Voice Conversations to ChatGPT at Scale

OpenAI has released GPT-Live, a new voice model designed to enable continuous, real-time conversation on ChatGPT. Unlike traditional voice interfaces that break the flow with alternating turns, GPT-Live uses full-duplex audio processing to allow simultaneous listening and speaking. This means the system can make brief acknowledgments, interruptions, and even web searches without disrupting the conversation.

At launch, GPT-Live-1 is available for both paid ChatGPT plans and free users, with GPT-Live-1 mini offering a more lightweight option. The model separates the real-time interaction layer from more complex tasks, which are handled by a separate backend model, initially identified as GPT-5.5. This division of labor allows GPT-Live to maintain fluid conversation dynamics while delegating demanding work to a more powerful model.

Currently, GPT-Live supports web search, memory, and visual widgets within ChatGPT Voice, but video or screen sharing is not yet supported. OpenAI has also emphasized the safety of the new voice system, adding safeguards to ensure appropriate responses and the ability to end conversations if necessary. While GPT-Live is currently a ChatGPT product, its potential for enterprise applications lies in providing hands-free interaction while accessing research, organizational context, or complex assistance.

However, organizations interested in implementing voice AI workflows will need to monitor for API availability and further details on model delegation as the technology develops.