OpenAI launched a new generation ChatGPT voice model
Wednesday, along with a more than three-minute video featuring three older women who explain how the feature works.
The company's vision is to enable natural human–AI interaction, create
a world where collaborating with artificial intelligence (AI) feels as fluid and responsive as working with another person, while reasoning and complex task execution take place in the background.
The three woman take turns explaining in a video that GPT-Live is built on a full-duplex
architecture. This means the voice model can listen and speak at the same time.
During conversations, GPT-Live can pay attention with phrases such as “mhmm” or “yeah,” and
engage in quick back-and-forth banter, or just remain quiet when the user needs a moment to think.
advertisement
advertisement
Each woman in the video talks with GPT-Live as if the artificial assistant is a
friend. The video opens with the first woman, S.J., sitting at a table knitting, talking in a heavy U.S. east coast accent to ChatGPT on speaker phone on the table.
"I'm making a sweater for
my grandson, but I don't want the needles to be too big, what do you think?" she asks, while knitting.
"If you go up [a size], it will get very baggy," GPT-Live says.
"Although
baggy is very popular now," the woman knitting says.
"It sure is," GPT-Live says.
"I know, right," the woman says.
The launch of OpenAI’s GPT-Live voice models will accelerate the value of intent data available to advertisers across the OpenAI Ads platform, despite ads today not available to
interact with spoken words.
OpenAI’s strict ad guidelines state that ads cannot influence chatbot responses or be spoken to directly by the AI, but the shift to a highly
natural, full-duplex voice interface will drastically change how users interact with the app.
GPT-Live is the company's smartest voice model yet, according to OpenAI. For questions that require
web search, deeper reasoning, or more complex work, it delegates the task of finding information to its latest AI frontier model behind the scenes and brings the result back into the conversation when
it has all the information needed.
If the model must go out to other
models and resources to find more information, GPT-Live can keep talking with the user to maintain the flow of the conversation.
The technology is built on cascaded voice systems that rely on
a series of models that act one after another to process each turn in conversation.
The original ChatGPT Voice strung three models
together, whereas this approach enables OpenAI to talk to frontier AI models for the first time.
This complexity came at a cost. Information could be lost across models, and responses were
slow and stilted. That changed with this version.
At launch, GPT-Live will use GPT-5.5 in the background. As OpenAI releases new
frontier models, it will continuously update the model used by GPT-Live.
OpenAI said the model can think and reason at the same time. GPT-Live has begun to roll out in two
versions, GPT-Live-1 and GPT-Live-1 mini, to ChatGPT users globally today. The plan is to bring the voice models to the API soon.