OpenAI today released novel conversational models called GPT-Live-1 and GPT-Live-1 mini, claiming they sound more natural and are better at making turns. These are full duplex models, which means they can speak and listen at the same time, allowing users to interrupt naturally and providing features such as live translation.
The company is also replacing its current advanced voice mode in ChatGPT with GPT-Live-1 mini by default. Paid tier users will be able to access the larger GPT-Live-1 model. The previous model combined a speech-to-text model for transcribing speech, a enormous language model for generating responses, and a text-to-speech model for delivering the final response.
The company said at a press conference that the novel models address issues such as users being interrupted while talking and not having enough intelligence to answer questions. The novel OpenAI models will query the latest text-based models, such as GPT-5.5, for search, inference, or agenting capabilities while continuing the conversation.
OpenAI also showed that the model can remain hushed for a long time and absorb the context of the conversation until it is triggered. Additionally, because the novel voice mode has access to newer GPT models, it can also present certain information in a visual format. Other startups like Monogram, which raised $40 million in seed funding from DST and Lux Capitalthey also rely on visual responses to boost the interactivity of assistants.
The company says the novel voice mode in ChatGPT is designed for longer conversations. During the briefing, ChatGPT Voice product manager Atty Eleti said he had 30 to 40 minutes of conversations using the voice feature during his walks.
OpenAI believes that voice can be the primary computing interface for performing elaborate work. Reports suggest that it may launch a pair of earbuds with AI features this year. However, no information has been provided about the hardware products.
“Over time, we believe this will also unlock the ability to use voice as the primary interface for processing data and managing the increasingly complex, long-term work of agents. We believe that voice can become the future interface for all kinds of work,” Eleti said.
Over the past few years, OpenAI has been working on improving voice features to make ChatGPT’s voice mode sound more natural. The company said that more than 150 million people talk to ChatGPT using features such as voice and dictation.
Rivals are also trying to make assistants more expressive.
Both Apple and Amazon have updated their assistants to provide more conversation and better context support. Startups like Sesame, founded by Oculus co-founder Brendan Iribe and Ankit Kumar, have also launched AI assistants that enable more natural conversation while performing background tasks.
OpenAI is moving in the same direction, aiming to allow users to talk to their assistant for longer periods of time without using their hands. Despite claims that the novel voice mode sounds more natural, the company stressed that it has no intention of making it an AI companion. It noted that the novel models have built-in safeguards to ensure teens receive age-appropriate responses and provide resources if the conversation turns to topics such as self-harm.
The novel voice mode still needs work. During a demonstration when the company showed off its live Hindi translation feature, the assistant had a bulky American accent and spoke in Hindi that sounded unnatural and had a slightly bookish tone. The company said the novel mode is optimized for “most languages in use,” but did not specify which ones.