source: techcrunch ai: openai releases new voice models for more natural live conversations
level: technical
openai today released two new conversational models, gpt-live-1 and gpt-live-1 mini. these are full-duplex models, meaning they can speak and listen at the same time. users can interrupt naturally, and features like live translation are possible. the company is replacing its current advanced voice mode in chatgpt with gpt-live-1 mini by default. paid users can access the larger gpt-live-1 model.
the previous voice system used separate speech-to-text, language model, and text-to-speech components. the new models solve issues like interrupting users and lacking intelligence for complex questions. they can send queries to newer text models like gpt-5.5 for search or reasoning while keeping the conversation going. the model can also stay silent and absorb context until called upon, and present information visually when needed.
openai sees voice as a potential primary interface for complex work. product lead atty eleti noted having 30- to 40-minute conversations during walks. over 150 million people already use chatgpt voice and dictation. rivals like apple, amazon, and startups are also making assistants more conversational. despite aiming for natural sound, openai stressed it is not building an ai companion. safeguards provide age-appropriate responses and resources for sensitive topics. a demo showed a heavy american accent in hindi translation, indicating room for improvement.
why it matters: full-duplex voice models enable more fluid human-computer interaction, which could change how people use ai for hands-free, long-duration tasks.
source: techcrunch ai: openai releases new voice models for more natural live conversations