OpenAI Launches GPT-Live Voice Model Enabling Real-Time Conversational Interaction

OpenAI has introduced GPT-Live, a new generation of voice models designed to enhance the voice chat experience in ChatGPT. This system allows continuous and natural voice communication without waiting for the model to finish speaking. Users can interrupt the model mid-response, add information during the conversation, or ask it to listen silently. The key innovation is based on Full-Duplex architecture, enabling simultaneous listening and speaking, unlike previous voice systems that followed strict turn-taking rules.

GPT-Live can respond in real time, pause, resume, or activate tools as the conversation evolves. OpenAI is releasing two versions: GPT-Live-1 for paid subscribers and GPT-Live-1 mini for free users. For more complex tasks like web searches or deep analysis, GPT-Live can run the flagship GPT-5.5 model in the background while maintaining conversational flow. Users can also select different thinking levels, from quick answers to more in-depth analysis.

The upgrade includes visual cards during voice chats, providing information on weather, stocks, and sports both audibly and on screen. OpenAI emphasizes integrated real-time safety mechanisms, including blocks against voice imitation of real people, adjustments for younger users, and parental controls. The new models are gradually rolling out worldwide, with GPT-Live-1 set as the default for Plus, Pro, and Go subscribers, and GPT-Live-1 mini available to free users.