OpenAI's GPT-Live-1 voice model speaks and listens simultaneously

OpenAI has overhauled ChatGPT’s voice mode with GPT-Live-1, a new model designed for natural, continuous conversation. Unlike the previous turn-based system, GPT-Live-1 is full-duplex, meaning it can speak and listen at the same time. It interrupts users less and waits when they pause mid-sentence, creating a flow closer to human dialogue. The model also introduces real-time translation—translating speech as you talk instead of waiting for pauses. When faced with complex queries, GPT-Live-1 automatically delegates to OpenAI’s best text models, such as GPT-5.5, for reasoning or web search, then quickly transitions back to voice. Additional new features include AI-generated visuals for contextual topics like stock prices, weather forecasts, or sports scores, and the ability to ask ChatGPT to remain silent until called upon, with acknowledgment phrases like “mhmm” or “got it.”

Safety has been a key focus. OpenAI has built in safeguards to steer conversations away from harmful responses and can end chats entirely in “higher-risk” situations, a move likely in response to lawsuits alleging that ChatGPT fueled delusions and harmed mental health. The model also provides expert-vetted crisis helpline support for self-harm discussions and delivers age-appropriate responses for teens. GPT-Live-1 is rolling out across iOS, Android, and the web, powering ChatGPT Voice for Go, Plus, and Pro subscribers. A smaller, more efficient GPT-Live-1 Mini will be the default for free users, ensuring broad accessibility while prioritizing performance for paid tiers.