ChatGPT's New Voice Model: GPT-Live-1 Mimics Human Talk

ChatGPT is testing a new voice model called GPT-Live-1, designed to mimic human conversation. This advanced model reduces interruptions and waits if a user pauses briefly while speaking. Here's the thing: The system can now speak and listen at the same time. This creates a more fluid conversational experience. What's interesting is that GPT-Live-1 can automatically transfer complex queries to powerful text models, like GPT-5.5, then quickly return with results. The update also enhances the user experience with visual data displays for things like weather or sports results. Older voice models often lost natural flow. This new model is "fully bidirectional," processing audio continuously. It even allows for real-time interpretation during speech. Users can also mute the assistant, and it will respond with vocal cues like "mm-hmm" while listening. OpenAI has added new safety systems to prevent harmful responses. This matters because it could make interacting with AI feel much more natural and efficient.