OpenAI released two new conversational voice models on July 8, 2026 — GPT-Live-1 and GPT-Live-1 mini — designed to handle more natural back-and-forth dialogue, including the ability to speak and listen simultaneously.
The models are described as full-duplex, meaning they can process incoming speech while generating a response. This enables users to interrupt naturally mid-sentence and supports features like live translation. OpenAI is replacing the current Advanced Voice Mode in ChatGPT with GPT-Live-1 mini as the default, while paid-tier users will have access to the larger GPT-Live-1 model.
Unlike the previous system — which chained together separate speech-to-text, language model, and text-to-speech components — the new models connect to OpenAI’s latest text models, including GPT-5.5, for search, reasoning, and agentic tasks while continuing the conversation. The models can also remain silent during a conversation and absorb context until called upon, and can present some information in a visual format.
OpenAI said the new voice mode is designed for longer conversations. Atty Eleti, ChatGPT Voice’s product lead, said during a press briefing that he has held 30- to 40-minute conversations with the feature during walks. The company stated that more than 150 million people currently use ChatGPT’s voice and dictation features.
OpenAI framed the development as a step toward voice becoming a primary interface for complex and long-running tasks. “We think voice can be the future interface to all kinds of work,” Eleti said. The company did not announce any hardware products, though reports have suggested it could release AI-capable earbuds.
The release is not without limitations. During a live demo of the Hindi translation feature, the assistant spoke with a heavy American accent and unnatural phrasing. OpenAI said the mode is optimized for “most spoken languages” but did not specify which ones. The company also noted that the new models include safeguards for age-appropriate responses to teens and for conversations involving topics like self-harm.
Source: TechCrunch