Introducing NVIDIA Cosmos 3: The Open Model That Thinks, Generates, and Acts

Hinzugefügt: 2026-06-06

[00:00:01]The real world is infinite and unpredictable. [music] Physical AI needs data, but real-world data [music] is impossible to scale.

[00:00:10]For physical AI, compute is data.

[00:00:14]This is cosmos, an open frontier omnimodel for physical AI built on a new mixture of transformers architecture.

[00:00:22]Pixels, action, sound, and language flow into the autoregressive transformer, >> [music] >> which reasons, plans, and instructs the diffusion transformer, which generates what comes next.

[00:00:34]Developers post-train cosmos across embodiments and use cases.

[00:00:39]As a VLM, cosmos watches the physical world, understands [music] what's happening, describing scenes, and flagging what matters.

[00:00:49]As a world model, cosmos generates physics-accurate synthetic video from an image, text, or video.

[00:00:58]As a simulator, cosmos closes the loop for policy training and evaluation. And as [music] the foundation of NVIDIA Omnidreams, an action-conditioned world model, cosmos predicts the future frame by frame.

[00:01:12]Post-train cosmos, and it becomes a world action model, perceiving, reasoning, planning, generating actions for robots of every kind, for everything that moves.

[00:01:30]A new kind of data, a new kind of teacher generated by computer.

[00:01:37]Cosmos, the foundation for developers of the age of physical [music] AI.