OpenAI has launched GPT-Live, a major upgrade to its voice AI. Unveiled on July 8, the model enables true full-duplex conversation-it can listen and speak at the same time while handling multiple background tasks.
The system uses a continuous processing architecture, replacing older, slower sequential models. This allows it to manage interruptions and pauses naturally. For complex queries requiring deep reasoning or web research, GPT-Live delegates tasks to background models like GPT-5.5, delivering real-time information such as flight statuses and stock prices without breaking the conversation flow.
User evaluations show strong preference for GPT-Live's fluid, human-like turn-taking compared to previous models.
OpenAI is releasing two tiers: GPT-Live-1 for paying subscribers and GPT-Live-1 mini for free users. An API is planned for developer integration. The service is available on iOS, Android, and the web.