OpenAI launches GPT-Live, introduces more natural, real-time voice conversations in ChatGPT

New Delhi: OpenAI unveiled GPT-Live, a new generation of voice models designed to make conversations with ChatGPT more natural and human-like. The company said the new technology powers an upgraded ChatGPT Voice experience with faster interactions, improved listening, smarter responses and richer visual capabilities. GPT-Live introduces a full-duplex architecture, allowing the AI to listen and speak simultaneously instead of waiting for users to finish talking before responding. This enables interruptions, quicker back-and-forth exchanges, and more fluid conversations that resemble human interactions. Unlike previous voice systems, GPT-Live can continue a conversation while delegating complex tasks such as web searches or deeper reasoning to OpenAI's latest frontier model in the background. At launch, GPT-Live uses GPT-5.5 for these advanced capabilities and will automatically adopt newer frontier models as they become available. The company is rolling out GPT-Live-1 and GPT-Live-1 mini globally across ChatGPT on iOS, Android, and the web. GPT-Live-1 will become the default voice model for Go, Plus, and Pro subscribers, while Free users will receive GPT-Live-1 mini. OpenAI also plans to make the models available through its API in the future. According to OpenAI, GPT-Live acknowledges users during conversations with natural cues such as "mhmm" or "got it", waits patiently when users pause to think, and is less likely to interrupt because of background noise. The company has also remastered ChatGPT's nine built-in voices for the new experience, making conversations sound more expressive and responsive. ChatGPT Voice can now access GPT-5.5 for more advanced reasoning and web searches. Users can choose between Instant, Medium, and High reasoning modes depending on whether they prioritise speed or deeper analysis. Voice conversations also support visual cards for topics such as weather, stocks and sports, alongside existing capabilities including memory, search, images and file uploads. OpenAI said GPT-Live has been trained with additional safeguards tailored specifically for voice interactions. The company expanded its testing to cover risks including self-harm, emotional dependence, violence and inappropriate content, while adding protections for teenagers through parental controls and age-appropriate behaviour. The model is also designed to prevent voice impersonation by limiting interactions to a predefined set of voices. While GPT-Live currently does not support voice conversations with video or screen sharing, OpenAI said these features are under development. The company added that it will continue refining the model based on real-world usage and expand support across more languages over time.