OpenAI Launches GPT-Live to Enhance ChatGPT Voice Conversations

OpenAI announced on Wednesday the launch of GPT-Live, a new series of voice models intended to create more natural conversations with artificial intelligence. The models are now integrated into ChatGPT Voice and are being rolled out globally to users on iOS, Android, and the web. Two versions are available: GPT-Live-1 for paid subscribers and GPT-Live-1 mini for free users.

The new technology utilizes a full-duplex architecture, which enables it to process user input continuously while generating its own speech. This allows the AI to listen and speak at the same time, engage in rapid exchanges, and use conversational affirmations like "mhmm." This approach differs from previous turn-based systems that required a user to finish speaking before the AI could respond.

According to OpenAI, GPT-Live can delegate more complex tasks, such as web searches or deeper reasoning, to powerful frontier models running in the background. At launch, it will use GPT-5.5 for these queries, with plans to upgrade the underlying model as new technology becomes available. The updated ChatGPT Voice experience also includes visual card responses for topics like weather and stocks, and users can select different reasoning levels for answers.

Ad

In head-to-head evaluations, both GPT-Live models were reportedly preferred over the previous Advanced Voice Mode in conversations lasting five to ten minutes. The company states that over 150 million people use its existing ChatGPT Voice and Dictation features on a weekly basis. OpenAI plans to make the new models available to developers through its API in the near future.

At launch, the GPT-Live integration does not support voice with video or screen sharing within ChatGPT. OpenAI has stated it is working to add these capabilities. Users can still access previous versions of ChatGPT Voice where those features are available.