OpenAI unveils real-time Voice AI models under GPT-Live family
09 Jul 2026, 02:28 PMThe models are built to handle simultaneous speech input and output, reducing delays and making interactions feel more fluid.
OpenAI has introduced GPT-Live, a new suite of voice AI models designed to listen and respond at the same time, enabling more natural, real-time conversations between users and AI systems.
The two variants: GPT-Live-1 and GPT-Live-1 mini will begin rolling out to users worldwide from Wednesday, said OpenAI in a blog.
The models are built to handle simultaneous speech input and output, reducing delays and making interactions feel more fluid, it added.
The Sam Altman-led AI company also noted that GPT-Live builds on the safety features of OpenAI's latest models and includes additional training for key risk areas, along with new safeguards designed specifically for voice-based interactions.
The launch follows OpenAI’s introduction of three audio-focused models for developers in May. Those models were aimed at helping developers build voice-enabled AI agents that can hold more natural conversations and perform tasks instantly during interactions.
Last month, OpenAI said that it has confidentially filed for its initial public offering (IPO), shortly after rival Anthropic also initiated plans to go public.



