NVIDIA open world foundation model (omni-model) for physical AI. Combines vision reasoning, multimodal generation and robot action prediction.
Parameters
65B (Super) / 16B (Nano)
parameters
Release date
31 May 2026
Access:APIDownloadHostedDeployment:☁ Cloud💻 Local📱 On-device
Overview
Applications
Access & deployment
APIDownloadHosted
CloudLocalOn-device
Weights: Open weights
Key parameters
🧩 Parameters: 65B (Super) / 16B (Nano)
✓ Fine-tuning
📥 Input: text, image, video, audio…
Robotics
Robot controlRobot manipulationBimanual manipulationEmbodied task planningScene understandingSpatial reasoningSpatial predictionEnvironment modelingVisual grounding
Technical specification
Parameters
65B (Super) / 16B (Nano)
parameters
License
OpenMDW 1.1 (Linux Foundation)
Features:✓ Fine-tuning
Modalities
⬇ Input
textimagevideoaudiorobot_sensorsrobot_state_data
⬆ Output
textimagevideoaudiorobot_actionsrobot_commandsmotion_trajectories
Capabilities and applications
Native model capabilities
Synthetic data generation
Generating synthetic datasets that preserve the statistical properties of the original — used for model training, testing, and privacy protection.
Category: structured_generation
Reasoning
Category: reasoning
Video Understanding
Category: video
Multimodal understanding
Category: multimodal
Planning
Category: planning
Robotics
Robot controlRobot manipulationBimanual manipulationEmbodied task planningScene understandingSpatial reasoningSpatial predictionEnvironment modelingVisual grounding
Application domains
Technical architecture
Core Architecture
Model Form
Sources and related pages
6 sources
WebNVIDIA Cosmos — Physical AI with World Foundation ModelsBlogHow Cosmos 3 Helps Physical AI Think Before It Acts (NVIDIA Blog)RepoCosmos3 collection on Hugging FaceReponvidia/Cosmos on GitHubReportCosmos 3 Technical ReportWebNVIDIA Launches Cosmos 3, the Open Frontier Foundation Model for Physical AI
