OpenAI has unveiled GPT-Live, a cutting-edge voice AI system designed to enhance conversational fluidity by enabling simultaneous speaking and listening capabilities. This latest innovation is based on a full-duplex architecture, allowing the AI to interject with natural conversational cues such as “mhmm” or “yeah,” thereby facilitating more seamless and rapid dialogue without the interruptions caused by lengthy pauses.
For tasks that involve more intricate requests requiring web searches or advanced reasoning, GPT-Live deftly delegates these to a more robust AI model that operates in the background, all while maintaining the ongoing conversation. At its introduction, these complex tasks are supported by the GPT-5.5 model, with plans to integrate support for newer models in future updates.
The announcement of GPT-Live coincides with OpenAI’s confirmation that its latest GPT-5.6 model series will soon be available to the public, pending a comprehensive cybersecurity review. This series includes the prominent Sol model, alongside the Terra and Luna variants, marking a significant step forward in AI development.
OpenAI has already started the global rollout of two versions of the new voice models, namely GPT-Live-1 and GPT-Live-1 mini, to ChatGPT users worldwide. Additionally, the company is set to offer GPT-Live through its API, thereby empowering developers and businesses to integrate real-time voice AI capabilities into their applications.