OpenAI launches GPT-Live, a new voice model for more natural chats

OpenAI has released the GPT-Live series, a new full-duplex voice model that can listen and speak at the same time, making conversations feel more natural. The model comes in two versions: GPT-Live-1 and GPT-Live-1 mini, and is available to all ChatGPT users now. It can decide to speak, pause, or interrupt multiple times per second, creating a more human-like chat experience.

For complex tasks, GPT-Live splits voice chat from deep thinking: it keeps the conversation flowing while sending tasks like web search or reasoning to the GPT-5.5 series models. OpenAI built a new evaluation system focusing on comfort and flow, and found that GPT-Live-1 series scored better than the previous Advanced Voice Mode in user preferences.

Over 150 million people use ChatGPT's voice features every week, for things like hands-free help, language practice, bedtime stories, and casual chats. The new model also shows visual cards for weather, stocks, and sports during voice conversations, and still supports search, memory, and file uploads. In safety tests, GPT-Live performed as well or better than Advanced Voice Mode on topics like self-harm, emotional dependency, and violence.