Why AI Developers Are Turning to World Models
Vídeo em Inglês, publicado pelo canal Practical AI.
Assistir ao vídeo original no YouTube
Assuntos: Inteligência artificial. Publicado no YouTube em 2026-09-29.
Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.
About NVIDIA Cosmos
NVIDIA Cosmos is an open platform for building world models for physical AI, such as robots and self-driving cars. It includes world foundation models, open datasets, training and optimization frameworks and data processing tools. The current family, Cosmos 3, is built on a Mixture-of-Transformers architecture and can reason over video, act as a physics-based world simulator, and generate data from text, images, video, sound and actions, according to NVIDIA.
Cosmos 3 comes in three sizes: Super (64 billion parameters), Nano (16 billion) and Edge (4 billion). The models are released under the OpenMDW 1.1 license from the Linux Foundation. The guest in the video, according to its description, is Ming-Yu Liu, vice president of Cosmos Lab at NVIDIA.
Fact sheet
- Platform: NVIDIA Cosmos
- Current model family: Cosmos 3 (Super 64B, Nano 16B, Edge 4B)
- Architecture: Mixture-of-Transformers
- License: OpenMDW 1.1 (Linux Foundation)
- Uses: Robot learning, autonomous vehicle training, video analytics agents
- Official page: NVIDIA Cosmos – official page
Read also
- I Tested LTX-2.5 for Physical AI — Robotics, World Models & Synthetic Data
- AMD Acquires World Labs | The Rise of Spatial AI Models | Uplatz
- Tesla Optimus Gen 3 Revealed — The Humanoid Robot That Could Change Everything!
Sources and references: NVIDIA Cosmos – official page · YouTube description of the video by the Practical AI channel.