Nvidia introduces Cosmos 3 Edge, a 4-billion-parameter model that enables robots and vision AI agents to reason, and generate actions locally.
With Cosmos 3 Edge, Nvidia is adding a new model to the Cosmos 3 family of open-world models. The model is built on Nvidia Nemotron, the open-weight models designed for advanced reasoning, coding, and complex workflows for AI agents. Unlike heavier variants, this version is lightweight enough to run on edge hardware, enabling robots to understand their environment and determine actions in real time without a cloud connection.
Developers can use the open Cosmos framework to adapt the model to specific robots, vehicles, sensors, and environments in about a day. Cosmos 3 Edge is deployable on RTX GPUs, DGX systems, and Jetson, including the newly announced T2000 and T3000 modules.
In addition, NVIDIA is introducing new Metropolis libraries and skills. These help developers use coding agents to build, train, and manage video intelligence systems with Cosmos at least six times faster.
Cosmos coalition expands to Japan
At the same time, NVIDIA is expanding the Cosmos Coalition to Japan. This brings together world model builders, AI developers, and physical AI leaders to further develop open world models. Among the companies planning to join are FANUC, Fujitsu, Hitachi, Kawasaki Heavy Industries, Kubota, Yaskawa Electric, SoftBank, Sony, NEC, Honda R&D, and Preferred Networks.
Members can contribute to and build upon the Cosmos platform, which includes open models, data curation libraries, datasets, and frameworks. The resulting world models are intended to help Japanese companies test and optimize physical AI systems before they are rolled out.
Tip: NVIDIA is working on a more affordable Blackwell AI chip