OpenAI said GPT Live separates voice interaction from advanced processing tasks.
GPT Live can listen and speak at the same time
Photo Credit: OpenAI
OpenAI has upgraded ChatGPT Voice with a new GPT Live system that brings more natural conversations, simultaneous listening and speaking, and smarter responses powered by its latest AI models. The update introduces GPT Live 1 and GPT Live 1 mini, replacing the previous voice experience for users across different subscription tiers. The rollout also extends to ChatGPT on iOS, Android, the web, and CarPlay. OpenAI says the new voice mode supports richer interactions, visual responses, and improved handling of longer conversations.
According to an OpenAI press release, the company has introduced GPT Live, a new family of voice models that now powers ChatGPT Voice. The update focuses on making spoken interactions feel more fluid while allowing the assistant to continue handling complex requests in the background. OpenAI has begun making GPT Live 1 and GPT Live 1 mini available to ChatGPT users globally, with API availability expected in the future.
The new voice system uses a full-duplex architecture that allows it to listen and respond simultaneously. Instead of waiting for a speaker to finish, the model continuously processes incoming audio and decides whether to reply, pause, continue listening, interrupt when appropriate, or use another tool. This approach enables faster conversations and also supports features such as live translation.
OpenAI said GPT Live separates voice interaction from advanced processing tasks. When a request needs web search, deeper reasoning, or agentic capabilities, the voice model hands the task to another AI model while continuing the conversation. The initial release uses GPT 5.5 as its underlying reasoning model, although OpenAI said future frontier models will replace it over time.
The company said GPT Live replaces earlier voice technologies that depended on sequential speech-to-text, language, and text-to-speech models. It also succeeds the turn-based Advanced Voice Mode, which required users to stop speaking before receiving a response and could misinterpret pauses or background sounds as the end of a conversation.
OpenAI said human evaluations showed GPT Live 1 and GPT Live 1 mini performed better than Advanced Voice Mode in overall conversation quality. The tests measured conversational flow, interruptions, turn-taking, naturalness, and user preference. The company also reported gains in GPQA for scientific reasoning, BrowseComp for agentic web search, and its internal voice telecom benchmark.
The updated ChatGPT Voice experience enables more natural conversations by recognising pauses without unnecessary interruptions. Users can interrupt the assistant, ask it to slow down, or request that it remain silent until needed. The model can also acknowledge that it is listening with brief verbal responses. OpenAI has also updated the nine built-in ChatGPT voices for GPT Live.
The new system can access OpenAI's latest frontier models to produce more capable responses. Users can choose between Instant, Medium, and High reasoning modes depending on whether they prefer faster replies or more extensive reasoning. ChatGPT Voice can also display visual cards for topics including weather, stocks, and sports while supporting search, memory, image handling, and file uploads.
OpenAI said the upgraded voice mode performs better in noisy environments by focusing on the user's speech. It can also wait through longer pauses instead of responding immediately, making conversations feel less rigid than before.
The company said GPT Live includes dedicated safety measures designed for voice interactions. It expanded testing with audio-based evaluations and synthetic scenarios covering areas such as self-harm, psychosis, mania, violence, sexual content, and emotional reliance on AI. Internal teams also carried out red team testing to identify risks specific to spoken conversations.
OpenAI added that GPT Live can intervene during conversations if it detects potentially unsafe responses. Depending on the situation, it may guide the conversation towards a safer reply, provide additional safety information, or end the interaction in higher-risk cases. Voice conversations involving self-harm can also present expert-reviewed crisis helpline resources. The system includes age-appropriate behaviour for teenagers, while parental controls allow parents to manage access and receive notifications in certain high-risk situations.
The company said it will continue monitoring emotional reliance after launch to improve its safeguards over time. It also stated that GPT Live is designed to use predefined ChatGPT voices and includes protections that prevent it from imitating the voice of a real person.
The rollout covers ChatGPT on Android, iOS, the web, and Apple's CarPlay interface. GPT Live 1 becomes the default voice model for Go, Plus, and Pro subscribers, while GPT Live 1 mini is the default for Free users. OpenAI said the models have been optimised for many widely used languages, although some languages may still have accent or fluency limitations. The initial release does not include video calls or screen sharing in Voice mode, although OpenAI said those capabilities are in development.
Get your daily dose of tech news, reviews, and insights, in under 80 characters on Gadgets 360 Turbo. Connect with fellow tech lovers on our Forum. Follow us on X, Facebook, WhatsApp, Threads and Google News for instant updates. Catch all the action on our YouTube channel.