NVIDIA's open omni-model family for physical AI: world simulation, reasoning and action generation for robots and AVs.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Cosmos 3 is NVIDIA's family of open world foundation models for physical AI. NVIDIA describes it as an omni-model that understands and generates text, image, video, ambient sound and action, built on a Mixture-of-Transformers architecture. It is part of the wider NVIDIA Cosmos platform, which also includes datasets, data curation tools, and a training and serving framework.
The family comes in three sizes: Super (64B parameters, aimed at the highest physics accuracy and generation quality), Nano (16B, aimed at fast post-training and specialization on NVIDIA DGX Station), and Edge (4B, with a Nemotron-based reasoner, for real-time VLM and world-action model use on edge devices). Uses listed include vision-language reasoning and video analytics, serving as a backbone for robot world action models, controllable world simulation, and synthetic data generation for robots and autonomous vehicles.
Models, training scripts and datasets are released openly, and the Cosmos Framework supports inference and post-training. Production deployment is offered through NVIDIA NIM and TensorRT LLM. The models are optimized for NVIDIA hardware such as Blackwell GPUs, which is the main practical constraint for users. Exact license terms were not in the provided material.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Open-source physical AI foundation model combining text, image, video, audio and action to generate synthetic data for robots and autonomous vehicles. Nano and Super variants, with models, training scripts and datasets released.
Closest neighbours by meaning, not just by category.
Model · NVIDIA
NVIDIA's real-time generative world model for closed-loop autonomous vehicle simulation, built on Cosmos.
Model · Nvidia
NVIDIA's family of open models for agentic AI: reasoning, vision, speech, retrieval and safety models.
Model
Research framework generating simulation-ready 3D assets (rigid, deformable, articulated) with physical properties.
Model · NVIDIA
NVIDIA's open multimodal model (120B total, 12B active parameters) for image, video and document understanding.
Model · NVIDIA
NVIDIA's open-weight ~550B mixture-of-experts LLM with a hybrid Mamba-transformer design, aimed at agentic workloads.
Model · Hao AI Lab
Distilled MiniMax H3 reference-to-video-and-audio model from Hao AI Lab, generating in 8 steps instead of 50.
