OpenAI launched GPT-Live on Wednesday, replacing its Advanced Voice Mode with a full-duplex voice system that lets ChatGPT listen and speak simultaneously. The upgrade fundamentally changes how users interact with the AI, enabling conversations that mirror natural human dialogue rather than the turn-based exchanges of previous iterations.
Two models power the rollout: GPT-Live-1 and GPT-Live-1 mini. OpenAI is deploying both globally starting immediately across iOS, Android, and ChatGPT.com. GPT-Live-1 becomes the default voice model for paid subscribers.
Full-duplex audio represents a significant technical shift. Unlike the previous Advanced Voice Mode, which required users to finish speaking before ChatGPT responded, GPT-Live processes speech and generates replies in real time. Users can interrupt ChatGPT mid-response, ask follow-up questions without waiting for a complete answer, and experience flow closer to authentic conversation. The system handles overlapping speech naturally instead of treating it as a technical failure.
The mini variant targets lower-latency applications and resource-constrained devices, offering a lighter footprint while maintaining core capabilities. Both versions benefit from OpenAI's latest training data and architectural improvements in audio understanding and generation.
This upgrade addresses a persistent friction point in voice AI. Previous systems felt stilted because they enforced artificial turn boundaries. GPT-Live eliminates that constraint. Users can say "wait, rephrase that" while ChatGPT is still speaking, or ask clarifying questions mid-explanation without resetting the conversation state.
The rollout timing targets ChatGPT's broadest user base. Paid subscribers get access first, with GPT-Live-1 mini available to free users on a delayed schedule. OpenAI has not disclosed specific latency metrics or bandwidth requirements, though the full-duplex architecture typically