Introducing Cosmos 3 Edge

NVIDIA's Cosmos 3 Edge is a powerful 4-billion-parameter world model that enables robots and vision AI to understand and act within their environments on edge devices. It features a unique architecture with two transformer towers that create a shared representation for vision, language, and action. This design allows the model to simulate possible futures, predict the visual results of an action, and generate real-time control commands. The compact model delivers high-throughput inference on memory-constrained systems and is released with post-training scripts for developers to build custom applications.