AI
OpenAI launches full-duplex voice models that let users interrupt naturally in real-time
The company's new GPT-Live models enable simultaneous speech and listening, allowing for more fluid conversations with better context handling and visual responses.
TechCrunch — AI
OpenAI has introduced a fresh generation of voice models designed to enable more fluid, natural-sounding conversations. The new GPT-Live-1 and GPT-Live-1 mini models use full-duplex technology, which lets the system speak and listen simultaneously—a key shift that allows users to interrupt without disrupting the flow of dialogue.
How the technology works#
Unlike the previous approach, which strung together separate speech-to-text, language model, and text-to-speech components in sequence, the new models operate as an integrated system. They connect to OpenAI's latest reasoning models, including GPT-5.5, while maintaining the conversation and can present responses in visual format alongside spoken words. The system can also remain silent while absorbing context, only responding when prompted.
Rollout and access#
ChatGPT's Advanced Voice Mode will shift to GPT-Live-1 mini for all users, while subscribers to paid tiers gain access to the more capable GPT-Live-1 variant. Product lead Atty Eleti noted during OpenAI's briefing that early tests have sustained conversations lasting 30–40 minutes, suggesting the models can handle extended interactions without fatigue.
Solving prior limitations#
OpenAI identified a pair of problems with its predecessor: the system would interrupt users mid-sentence and lacked sufficient intelligence to handle complex queries. The new architecture remedies both by delegating reasoning to more sophisticated models and enabling natural turn-taking in dialogue.
Competitive landscape and vision#
Rivals including Apple and Amazon have rolled out updates to improve conversational fluidity in their assistants, while startups like Sesame, backed by Oculus co-founder Brendan Iribe, have launched competing voice assistants. OpenAI describes voice as potentially becoming "the primary interface to computing" for managing complex, long-running tasks—a vision that aligns with reports of planned hardware like .
