Summary
OpenAI introduces the new ChatGPT Voice, powered by a model called GPT Live, in this German-language launch video. The company describes it as bidirectional and full duplex, the most capable voice model it has built so far, meaning it can listen while it speaks and can be interrupted naturally.
The demo follows a group of friends at a hair salon who talk over one another while asking for haircut advice and taking photos. ChatGPT Voice keeps up with the overlapping conversation and works out when it is being addressed. The presenters then show it handling several thoughts at once: planning a fall bicycle tour from Frankfurt to Warsaw within a daily distance and budget limit, searching the web for a contact at a university in Szczecin, and translating into Polish in real time while continuing to research.
The video also shows the assistant working on tasks beyond conversation. It adds an appointment to a calendar, tells a short story about a botanist and a moth to prompt discussion, and builds presentation slides on request. Together, the demos show a voice assistant moving from simple turn-taking toward real-time, multitasking help.
📺 Source: OpenAI · Published October 19, 2026
🏷️ Format: Keynote Launch
