OpenAI released a new voice model, GPT-Live-1, on July 8, 2026. The model is designed for full-duplex conversations, real-time translation, and improved user interaction across various platforms.

GPT-Live-1 is configured to interrupt users less frequently than prior models. The system will wait for users to continue speaking if they pause during a conversation. OpenAI research lead Kundan Kumar stated, "This is the smartest voice model yet."

The model automatically routes queries to text models, such as GPT-5.5, for tasks requiring reasoning or web searches. It also supplements conversations about topics like weather, stocks, and sports with AI-generated visuals. Users have the option to command ChatGPT Voice to cease talking until it is prompted.

Atty Eleti, an OpenAI product lead, said that GPT-Live-1 is "a full duplex model." "What it really means is that it can speak and listen at the same time," Eleti said. He added, "From the model side, it can process the stream of inputs and produce the stream of output continuously and simultaneously." The model acknowledges it is listening through phrases such as "mhmm," "yeah," and "got it."

GPT-Live-1 supports real-time translation while a user is speaking. The model includes safeguards to guide it away from harmful responses and to end chats in situations deemed higher-risk. It is also trained to offer expert-vetted crisis helpline support during discussions about self-harm and is designed to provide age-appropriate responses for teenagers.

GPT-Live-1 is accessible on iOS, Android, and web platforms. The model powers ChatGPT Voice services for Go, Plus, and Pro subscribers. A version named GPT-Live-1 mini serves as the default for free users.

OpenAI is facing lawsuits alleging ChatGPT fueled delusions and harmed the mental health of users. The development of GPT-Live-1 includes specific provisions for offering crisis helpline support for self-harm.