OpenAI's gpt-live-1 finally stops waiting its turn

OpenAI just made ChatGPT's voice mode less like a walkie-talkie and more like a person. GPT-Live-1, announced during a press briefing, uses full-duplex capability: it processes incoming speech and generates outgoing speech at the same time. You don't wait for it to finish before it hears you.

Kundan Kumar, OpenAI's research lead, called GPT-Live-1 the company's "smartest voice model" so far. The model kicks complex queries to text-based reasoning engines like GPT-5.5 when it needs to search the web or do deeper analysis, then switches back to voice. Less lag between asking and hearing.

"What it really means is that it can speak and listen at the same time," said Atty Eleti, OpenAI's product lead, during the briefing. "From the model side, it can process the stream of inputs and produce the stream of output continuously and simultaneously."

The new setup also supports real-time translation: ChatGPT can start translating a user's speech before they finish a sentence, which the older turn-based model could not handle. Users can also tell the voice mode to stay quiet until spoken to, another capability OpenAI said was not possible before. In listening mode, the model might respond with sounds like "mhmm" or "got it" to show it is following along.

OpenAI added safety guardrails that steer the model away from harmful responses or end conversations in "higher-risk" cases. The company faces multiple lawsuits claiming ChatGPT fueled delusions and harmed users' mental health. GPT-Live-1 is trained to provide "expert-vetted crisis helpline support" in conversations about self-harm and to give age-appropriate responses for teenage users.

GPT-Live-1 is rolling out across iOS, Android, and the web. It will power ChatGPT Voice for Go, Plus, and Pro subscribers. A smaller, more efficient version, GPT-Live-1 mini, will be the default voice experience for free users.