ChatGPT’s upgraded voice mode now interrupts less

OpenAI has rolled out a major upgrade to ChatGPT’s voice mode, introducing a new model designed to make conversations feel far more natural. The company says the updated system, called GPT-Live-1, is engineered to interrupt users less frequently and to pause patiently when someone stops speaking mid-sentence, mimicking the rhythm of real human dialogue.

During a press briefing, OpenAI research lead Kundan Kumar described GPT-Live-1 as the company’s “smartest voice model” to date. A key improvement is its ability to automatically route complex queries to the most capable text models, such as GPT-5.5, when deeper reasoning or web searches are required. This allows the system to shift seamlessly from researching a topic to discussing its findings without awkward delays. The model also enriches conversations about weather, stocks, and sports by generating on-the-fly visuals, like current sports scores or a weekly forecast.

The previous voice system operated on an older, turn-based framework that often produced inaccurate answers and struggled with natural conversation flow. “This is a full duplex model,” explained OpenAI product lead Atty Eleti. “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.”

This real-time capability unlocks new features like live translation. Instead of waiting for a user to finish speaking before translating, ChatGPT can now render speech into another language while the person is still talking. Users can also instruct the voice mode to remain silent until explicitly called upon, a request that was not previously supported. The system will acknowledge active listening with subtle cues like “mhmm,” “yeah,” or “got it.”

Openai has also integrated built-in safety measures to steer the model away from harmful responses and to terminate conversations entirely in “higher-risk” situations. This update comes as the company faces multiple lawsuits alleging that ChatGPT has fueled delusions and harmed users’ mental health. The model now offers expert-vetted crisis helpline support for discussions involving self-harm and provides age-appropriate responses tailored to teenage users.

The rollout of GPT-Live-1 is underway across iOS, Android, and the web. It will power ChatGPT Voice for Go, Plus, and Pro subscribers, while a streamlined version called GPT-Live-1 mini will serve as the default voice model for free users.