Summary:NVIDIA has introduced Cosmos 3, a new family of open-world AI models designed to accelerate Physical AI development for robotics, autonomous vehicles, and vision-based systems. The platform combines world modeling, simulation, and edge deployment capabilities in a unified framework.

 

 

---

August 2026 — NVIDIA has unveiled Cosmos 3, a new generation of open-world AI models aimed at accelerating the development of Physical AI, a rapidly growing field that powers robots, autonomous vehicles, and intelligent vision systems.

 

The Cosmos 3 family includes three models: Cosmos 3 Super (64B) for high-fidelity world modeling, Cosmos 3 Nano (16B) for efficient reasoning and AI post-training, and Cosmos 3 Edge (4B) for on-device vision reasoning and robot policy deployment. Together, they provide a unified AI foundation for simulation, perception, and decision-making.

 

Integrated with NVIDIA Omniverse and OpenUSD, Cosmos 3 enables developers to generate physically grounded training data, simulate real-world environments, and build AI systems capable of understanding and interacting with the physical world.

 

 

According to NVIDIA, Cosmos 3 achieved top rankings across multiple industry benchmarks, including world generation, robot policy learning, and vision understanding. The models are optimized for deployment on NVIDIA RTX GPUs, Jetson platforms, and edge AI systems.

 

Several major companies, including Doosan Robotics, LG Electronics, Samsung Electronics, Skild AI, and autonomous driving developers such as Xiaomi, have already adopted Cosmos 3 for Physical AI training.

 

As global competition intensifies in robotics AI, autonomous driving, world models, and embodied intelligence, NVIDIA Cosmos 3 is emerging as a key platform for the next generation of intelligent machines.