New NVIDIA Cosmos 3 Open Models for Interactive World Simulation

Judgment

Teams weighing open world models must decide whether to adapt Cosmos3-Nano-Transfer-Auto to their own sensor rig and distill it into Sim-Auto for streaming rollouts, accepting that closed-loop use still needs a simulation runtime and WSM renderer.

Basis

NVIDIA authors Pranjali Joshi, Rick Z, Yao Xu, Qianli Ma, Chen-Hsuan Lin, Ming-Yu Liu, Xiang Li, Zhengfei Kuang, Zhonghao Wang and kiui present four new open-weight Cosmos 3 checkpoints, all post-trained from Cosmos 3 Nano with 16B trainable parameters under OpenMDW1.1, which permits commercial and non-commercial use. Transfer-Auto generates synchronized video across up to 11 camera views plus joint RGB–LiDAR, using per-camera World Scenario Map controls and text prompts; Sim-Auto is its causally distilled streaming variant, generating small chunks in a few denoising steps from rolling history and updated driving controls. The mechanism is adaptation: for a new sensor rig, teams post-train Transfer-Auto first, then distill it into Sim-Auto; closed-loop interaction still requires a simulation runtime and WSM renderer to update controls.