NVIDIA announced four post-trained Cosmos 3 Nano checkpoints for interactive world simulation: Transfer-Auto, Sim-Auto, Sim-Transfer and Sim-Bimanual. The publisher describes 16 billion trainable parameters for each. Read the original announcement and project details.
Why it matters
Their intended tasks range from varying driving-scene weather and lighting to generating synchronized camera and LiDAR observations and predicting video from robot actions. NVIDIA describes them as starting points for research and adaptation, not substitutes for real-world safety validation.
What to know
The models use OpenMDW 1.1, which permits commercial and non-commercial use under its terms. Review the individual model cards, hardware requirements and license before deployment. We have not run or benchmarked the models.
Reported from linked primary sources. No model or product was independently tested.
