OpenAI releases GPT-Live-1, full-duplex voice model that speaks and listens simultaneously

OpenAI has released GPT-Live-1, a full-duplex voice model that can process audio input and generate speech output simultaneously. The model replaces ChatGPT's previous turn-based voice system, which OpenAI says struggled with conversational flow and accuracy.

Technical capabilities

"This is a full duplex model," said OpenAI product lead Atty Eleti during a press briefing. "What it really means is that it can speak and listen at the same time… from the model side, it can process the stream of inputs and produce the stream of output continuously and simultaneously."

GPT-Live-1 automatically routes queries to OpenAI's text models, including GPT-5.5, when reasoning or web search is required. According to OpenAI researcher lead Kundan Kumar, this allows faster transitions from research to verbal responses.

The model now supports real-time translation during speech, rather than waiting for pauses. Users can also command the model to remain silent until called upon, with the system acknowledging listening through phrases like "mhmm," "yeah," and "got it."

Deployment and pricing

GPT-Live-1 is rolling out across iOS, Android, and web platforms. The full model powers ChatGPT Voice for Go, Plus, and Pro subscribers. Free users will receive access to GPT-Live-mini-1, described as a smaller, more efficient version. Pricing details for the individual models were not disclosed.

Safety features

OpenAI claims the model includes built-in safeguards that either redirect harmful conversations or terminate chats in "higher-risk" situations. The company states the model is trained to provide "expert-vetted crisis helpline support" for self-harm discussions and "age-appropriate" responses for teenage users.

These safety measures come as OpenAI faces multiple lawsuits alleging ChatGPT contributed to user mental health issues and fueled delusions.

Additional features

The model generates AI visuals for weather, stocks, and sports queries, displaying information like scores and forecasts during voice conversations.

What this means

Full-duplex voice processing represents a shift from stateless, turn-based voice systems to continuous audio streaming. This architecture enables interruptions and real-time translation but requires different safety controls than text-based models. The automatic routing to GPT-5.5 suggests OpenAI is building a hybrid system where specialized models handle different cognitive tasks, rather than a single unified model. The two-tier deployment (full model for paid users, mini for free) indicates computational costs remain a constraint for real-time voice processing.