OpenAI launches new GPT-Live Voice models that enable simultaneous listening and speaking

OpenAI has launched a major upgrade to ChatGPT Voice Mode powered by GPT-Live, a new generation of real-time voice models that replaces earlier speech architectures. The update introduces a full duplex voice design, allowing ChatGPT to speak and listen at the same time instead of relying on the turn-based approach used until now, where the chatbot waited to detect a pause in the user’s speech and interpreted it as its moment to respond. GPT-Live-1 models can now process user input while generating speech, making conversations feel more natural and responsive.

Users can interrupt ChatGPT Voice at any moment, and the model will adapt without restarting or pausing the conversation. GPT-Live can also add subtle verbal acknowledgments like “okay” or “got it” when a user hesitates, stay silent when more thinking time is needed, slow down when asked, and better focus on speech in noisy environments. The final result is a more fluid conversation, somewhat similar to the now-iconic Scarlett Johansson-style voice from the first introduction of Voice Mode.

For more complex requests, GPT-Live can hand off tasks to advanced OpenAI systems like GPT-5.5 while the voice chat continues in the background. The update also adds visual cards for weather, sports, and stocks, while keeping support for image uploads, files, and web search. OpenAI has added new safeguards, including parental controls for ChatGPT Voice and parent alerts in certain teen safety cases. GPT-Live-1 is available for paid users, GPT-Live-1 mini is available for free users, and API access is planned for developers.