OpenAI launches GPT-Live-1 voice model with full duplex capabilities and improved interruption handling
OpenAI has introduced the GPT-Live-1 voice model, featuring full duplex capabilities, improved interruption handling, and real-time visual integration for subscribers across mobile and web platforms.
1. New Voice Model Capabilities
OpenAI has introduced GPT-Live-1, an upgraded voice model for ChatGPT designed to facilitate more natural, human-like interactions. Unlike the previous turn-based system, this "full duplex" model can process inputs and outputs simultaneously, allowing it to listen and speak at the same time. The model is engineered to interrupt users less frequently and can detect pauses to wait for a user to finish speaking. Additionally, it supports real-time translation and can be instructed to remain silent until prompted.
2. Enhanced Features and Integration
The GPT-Live-1 model is described by OpenAI research lead Kundan Kumar as the company’s smartest voice model to date. It automatically routes queries to advanced text models, such as GPT-5.5, to improve reasoning and web search capabilities. The system can also provide AI-generated visuals to supplement conversations regarding topics like sports scores, stock market data, and weather forecasts.
3. Safety and Availability
OpenAI has implemented new safety safeguards within the model to steer it away from harmful content and to provide expert-vetted crisis support for conversations involving self-harm. The company has also designed the model to offer age-appropriate responses for teen users. GPT-Live-1 is currently rolling out across iOS, Android, and the web. It will be available to Go, Plus, and Pro subscribers, while free users will have access to a more efficient version known as GPT-Live-1 mini.