OpenAI is updating its voice AI technology to match its text-based artificial intelligence. Two new versions, GPT-Live-1 and a smaller version, are now being released to everyone ChatGPT users.
The new model duo relies on the latest OpenAI models. You can choose from three different levels of intelligence, depending on how deep you want the answers. But the biggest update should be that the voice sounds more human and more expressive, Atty Eleti, ChatGPT’s voice product leader, told reporters.
One of the biggest changes under the hood is a new process called continuous collaboration. This framework allows AI to simultaneously receive information and generate output — so in the case of voice, it can “listen” and “speak” at the same time. This is a change from previous vocal methods that can only be answered after you finish speaking.
It’s useful if you want to translate live, simultaneously: You can speak in English and have ChatGPT translate what you’re saying into Spanish, Hindi or another language with a short delay.
Part of the new architecture can also assign or outsource functions to OpenAI boundary models. For example, it can answer a question you asked while researching another.
“When GPT-Live has to think hard about a question, it can transfer its thinking and complex work to GPT-5.5, which can do things in parallel, and this GPT-Live can still stay in conversation with the user,” said Kundan Kumar, who leads the GPT-Live model research. When it’s done, it “seamlessly weaves” the response into its response.
“This is exactly how people interact with each other,” Eleti added. “We’re keeping the conversation going while we’re thinking behind the scenes.”
It also uses OpenAI’s new AI design tech creating visual responses, if appropriate, “because sometimes the best response is shown, not spoken aloud,” said Eleti. Think weather reports, sports scores and other types of images.
This is an example of a graphic ChatGPT can do if you ask it to give you a weather report while in voice mode.
These new voice models are now available to all ChatGPT users, including free users and paying subscribers, on mobile apps and on the website. Don’t worry if you don’t see it right away; OpenAI says it may take a day or two to ensure everyone has access. This is different from what was expected on Thursday GPT-5.6 series download.
You’ll notice that one way the AI voice tries to mimic human speech is with full words — those “ums,” “ers” and “likes” we use when thinking our thoughts out loud. Those are new in this update. You can also interrupt it with a few delays. You can make it sit still and listen, it only responds when its name is called as a wake-up call. Unlike Siri and Alexa, however, ChatGPT will listen continuously until manually disabled.
OpenAI automatically takes you out of AI training in voice mode. Audio clips are stored for 30 days — so the AI has the context of previous conversations — and can be deleted.
Having an AI that feels more human comes with a risk, though. We have seen many cases (and cases) for the injury of anthropomorphizing AI it can be devastating, especially for people with mental health issues. OpenAI says the new models have increased protections and are more effective than previous models in “critical safety areas” including self-harm, mental illness, violence and sexual content.
