.png){kind=logo}
OpenAI introduced GPT-Live, a new generation of voice models designed to make conversations with AI feel more like a natural spoken exchange. The technology powers the latest ChatGPT Voice experience and is built to handle interruptions, pauses, and changes in conversation without forcing users to wait for one response to finish before speaking again.
The main change with GPT-Live is its full-duplex design. Instead of treating a voice conversation as a series of separate turns, the model can listen and speak continuously. This means users can interrupt ChatGPT while it is talking, change direction mid-sentence, or pause before continuing.
The system can decide whether to respond immediately or keep listening, which is intended to make conversations feel less like a voice command interface and more like a back-and-forth discussion.
OpenAI offers GPT-Live-1 for paid ChatGPT users, while free users get GPT-Live-1 mini. The models are rolling out globally across ChatGPT.com and the iOS and Android apps.
GPT-Live is not limited to spoken replies. Voice conversations remain part of a regular ChatGPT conversation, so users can follow the response in text and switch between typing and speaking.
The Live experience can also use features such as web search and memory where available. It can work with text and images in the same conversation and can display supported visual results. According to OpenAI, the current Live experience does not support video or screen sharing; those capabilities remain available through the older Advanced Voice option for eligible users.
OpenAI has also designed GPT-Live to work with backend models and tools. The voice model handles the conversation, while more involved tasks can be passed to a backend agent that reasons or uses connected tools.
For example, a voice assistant can continue talking with a user while a backend checks information or completes a task. The application remains responsible for permissions, confirmations, and executing custom actions.
Also Read: OpenAI Astra for Law with GPT-6 Astra: Features and Availability
The launch marks a shift in how OpenAI is approaching spoken interaction. The focus is no longer simply on converting speech into text and reading an answer aloud. GPT-Live is designed to manage the conversation itself, including timing, interruptions, and handoffs to other systems.
OpenAI also added GPT-Live to its API, giving developers options to build real-time voice applications through WebRTC, WebSockets, and telephony connections. The company added SynthID watermarking to supported audio generated through GPT-Live, allowing OpenAI-generated audio to carry provenance signals that can be detected by its verification tools.
