NVIDIA Releases Four Open-Weight Cosmos 3 Models for Interactive World Simulation
Image via huggingface.co
NVIDIA has expanded its Cosmos 3 world model platform with four new open-weight checkpoints — Cosmos3-Nano-Transfer-Auto, Cosmos3-Nano-Sim-Auto, Cosmos3-Nano-Sim-Transfer, and Cosmos3-Nano-Sim-Bimanual — each targeting a distinct use case in robotics or autonomous driving simulation. All four are post-trained variants of Cosmos 3 Nano with 16 billion trainable parameters and are released under the OpenMDW 1.1 license, permitting both commercial and non-commercial use.
The models address two core simulation challenges: generating synchronized multi-sensor data (RGB cameras and LiDAR) for autonomous driving research, and predicting video outcomes from robot or camera actions for robotics policy development. Capabilities include multi-view driving scene generation across up to 11 camera angles, streaming video prediction tied to ongoing actions, and converting physics-simulator controls plus text prompts into photorealistic video output. NVIDIA frames the release as an open foundation for researchers to adapt, benchmark, and build upon.