OpenAI has released new conversational models, GPT-Live-1 and GPT-Live-1 mini, claiming they sound more natural and can handle turn-taking better than previous versions. These full-duplex models can speak and listen simultaneously, allowing users to interrupt naturally and enabling features like live translation. The company is replacing its current Advanced Voice Mode in ChatGPT with GPT-Live-1 mini by default, while paid users get access to the larger GPT-Live-1 model. The old setup - a speech-to-text model, a large language model, and a text-to-speech model strung together - apparently had some issues, like interrupting users mid-sentence and lacking the intelligence to answer questions. The new models can send queries to GPT-5.5 for search, reasoning, or agentic tasks while keeping the conversation flowing.

OpenAI also demonstrated that the model can stay silent for long periods, absorbing context until called upon, and can present visual information thanks to access to newer GPT models. Competitors like Monogram, which raised $40 million from DST and Lux Capital, are also leaning into visual responses. During a press briefing, ChatGPT Voice product lead Atty Eleti claimed he's had 30- to 40-minute voice conversations during walks. OpenAI envisions voice as the primary interface for complex computing, and rumors suggest it might launch AI-powered earbuds this year, though the company stayed mum on hardware.

“Over time, we think this will also unlock the ability to use voice as a kind of primary interface to computing, and to manage increasingly complex long-running agentic work,” Eleti said. OpenAI has been refining voice features for years, noting that over 150 million people use Voice and Dictation in ChatGPT. Rivals like Apple and Amazon are also sprucing up their assistants, and startups like Sesame (co-founded by Oculus co-founder Brendan Iribe and Ankit Kumar) are launching more natural AI assistants. OpenAI insists it's not trying to create an AI companion, despite making voice mode more engaging, and has added safeguards for age-appropriate responses and resources for self-harm topics. However, the new mode still needs work: during a demo of live translation into Hindi, the assistant spoke with a heavy American accent and a bookish tone. OpenAI says the mode is optimized for “most spoken languages” but didn't specify which ones.