Cosmos 3
$ 0
Cosmos 3 is NVIDIA’s open world model for physical AI. It understands scenes, generates future video and predicts actions, helping teams train and test robots faster.
Available on backorder
Brain Score
2
Specifications and details:
| Nationality | US |
|---|---|
| Website | https://www.nvidia.com |
| Model type | Foundation Model |
| Manufacturer | NVIDIA |
| Release date | June 1, 2026 |
Description
Cosmos 3 is NVIDIA’s open world foundation model for physical AI. NVIDIA launched it at GTC Taipei in June 2026. It joins reasoning, video generation and action prediction in a single system. NVIDIA also calls it the first fully open omnimodel, because it handles text, images, video, sound and actions. That breadth makes it a shared base for robot and self-driving car teams.
Teams can use Cosmos 3 to simulate scenes and predict what happens next. They can also generate synthetic training data and build world action models that teach robots specific jobs. For example, the model can predict robot actions from video for arms such as Franka, UR and AgiBot. NVIDIA ships it in sizes, with Super for top accuracy, Nano for speed and Edge coming later. Developers can try it on build.nvidia.com or download the weights from Hugging Face. Meanwhile, NVIDIA formed a Cosmos Coalition with partners such as Skild AI and Agile Robots to grow open world models.
New! 2026 Humanoid
Robot Market Report
198 pages of exclusive insight from global robotics experts — uncover funding trends, technology challenges, leading manufacturers, supply chain shifts, and surveys and forecasts on future humanoid applications.
now Google DeepMind
Contact Humanoid.guide
Website: https://research.nvidia.com/labs/cosmos-lab/cosmos3/
