The artificial intelligence industry has made another qualitative leap. OpenAI has officially announced GPT-Live, a new generation of voice models that is already integrated into the "Conversation" mode in ChatGPT. This is not just an update, but a fundamental shift in how we interact with neural networks.
The key innovation is the full-duplex architecture. Unlike previous solutions, GPT-Live can listen and speak simultaneously. The model no longer waits for the user to stop talking before starting its response. It recognizes interjections, makes natural pauses, and, crucially, does not interrupt if the person is simply thinking. This makes the dialogue with AI as close to human interaction as possible.
The system architecture separates the processes of communication and task execution. While GPT-Live maintains an active dialogue, another model (initially GPT-5.5) operates in the background. It handles internet searches or performs complex calculations. This approach allows the AI to translate speech in real-time and resolve requests without interrupting the conversation.
According to internal test results, GPT-Live significantly outperforms previous versions in scientific reasoning and data analysis. Visual cards have also appeared in the ChatGPT interface — users can now instantly get information about the weather, stock prices, or sports results without leaving the dialogue.
OpenAI has released two versions: the flagship GPT-Live-1 and the lightweight GPT-Live-1 mini. Access to the main model is open to Plus and Pro subscribers. Users of the free version of ChatGPT can use the mini version. In the near future, the company plans to open an API for developers, which will undoubtedly lead to mass integration of the technology into third-party services.
Special attention has been paid to safety. New filters have been implemented in the system that block attempts to imitate other people's voices and restrict content related to violence or self-harm. Parental controls are provided for teenagers. This is a timely step, given the growing concerns about the use of voice deepfakes.
Expert opinion. GPT-Live is not just an improvement of a voice assistant. It is a transition to a new paradigm of human-machine interaction, where the line between talking to AI and a live interlocutor is blurred. We are witnessing not an evolution, but a revolution in the UX design of AI products.