OpenAI Launches GPT-Live Voice Models for Natural Interaction

OpenAI released two new conversational voice models, GPT-Live-1 and GPT-Live-1 mini, which replace the previous Advanced Voice Mode in ChatGPT. The models utilize a full-duplex architecture to enable simultaneous listening and speaking, reducing interruptions and allowing users to interrupt the AI naturally. GPT-Live-1 is available to Go, Plus, and Pro subscribers, while the mini version serves as the default for free users across iOS, Android, and the web.

To maintain fluid conversation, the system delegates complex reasoning, web searches, and agentic tasks to the GPT-5.5 text model in the background. This integration supports low-latency processing, live translation, and the presentation of rich visual cards for specific topics. The update also includes remastered versions of nine built-in voices and enhanced safety safeguards for voice-specific risks, though video and screen sharing are not yet supported.

During demonstrations, the live translation feature in Hindi exhibited an unnatural, bookish tone and a heavy American accent. Despite this, the company emphasized that the tool is intended for complex work and long-running agentic tasks rather than as a companion. The release aims to transition voice interaction into a primary interface for computing and complex workflow management.