OpenAI has unveiled GPT-Live, an innovative voice AI system designed to enhance the natural flow of conversations by enabling simultaneous speaking and listening. This advancement, based on a full-duplex architecture, allows the AI to provide responses while users are still speaking. By incorporating simple verbal acknowledgments like “mhmm” or “yeah,” GPT-Live facilitates smoother interactions without the need for extended pauses.
The system is adept at handling more intricate requests requiring web searches or advanced reasoning by delegating these tasks to a more robust AI model operating in the background, all while maintaining the ongoing dialogue with the user. Currently, these sophisticated tasks are managed by GPT-5.5, with OpenAI planning to introduce support for newer models in subsequent updates.
Following the announcement, OpenAI confirmed that the GPT-5.6 model series, which includes the flagship Sol model and variants Terra and Luna, will be released to the public once they pass further cybersecurity evaluations. This series is poised to enhance the capabilities of AI applications even further.
OpenAI has begun deploying two versions of its new voice models—GPT-Live-1 and GPT-Live-1 mini—making them accessible to ChatGPT users globally. Additionally, the company intends to offer GPT-Live via its API, providing developers and businesses an opportunity to integrate real-time voice AI capabilities into their applications.