NVIDIA Cosmos3-Edge, Nano, and Super Physical AI Models Now on Amazon SageMaker JumpStart
NVIDIA's Cosmos3-Edge, Cosmos3-Nano, and Cosmos3-Super models are now available on Amazon SageMaker JumpStart, expanding the portfolio of foundation models for AWS customers. These omnimodal world models are designed for physical AI applications, enabling the development of robots, autonomous vehicles, and vision AI capable of perception, reasoning, planning, and action. Each model offers specialized capabilities, from on-device control to high-fidelity world generation and simulation. Customers can deploy these models with a few clicks via the SageMaker JumpStart model catalog.
- →NVIDIA Cosmos 3 Family of Omnimodal World Models now on SageMaker JumpStart
- →Cosmos3-Edge model optimized for on-device robot control and real-time visual reasoning
- →Cosmos3-Nano model excels in physics-aware world generation and physical reasoning
- →Cosmos3-Super model provides high-fidelity world generation and simulation
Features (4) ›
- NVIDIA Cosmos 3 Family of Omnimodal World Models now on SageMaker JumpStart
NVIDIA's Cosmos3-Edge, Cosmos3-Nano, and Cosmos3-Super models are now available on Amazon SageMaker JumpStart. These open, frontier omnimodal world models are designed for physical AI, enabling customers to build robots, autonomous vehicles, and vision AI that perceive, reason, plan, and act.
- Cosmos3-Edge model optimized for on-device robot control and real-time visual reasoning
Cosmos3-Edge is a 4B-parameter omni-model engineered for on-device robot control and real-time visual reasoning on edge hardware. It operates at robot-control resolution (640×360), delivering real-time reasoning and generating 32 actions per inference at 15 Hz on NVIDIA Jetson Thor, supporting 256p and 480p video at 12–30 FPS.
- Cosmos3-Nano model excels in physics-aware world generation and physical reasoning
Cosmos3-Nano is a compact 16B-parameter omnimodal model that processes combinations of text, image, video, audio, and action trajectories to produce corresponding outputs. It enables robots and vision AI agents to reason using prior knowledge, physics understanding, and common sense, supporting chain-of-thought reasoning over text, images, and video up to 720p.
- Cosmos3-Super model provides high-fidelity world generation and simulation
Cosmos3-Super is the 64B-parameter model offering the highest-fidelity world generation and simulation in the Cosmos 3 family. It jointly processes and generates language, images, video, audio, and action sequences within a unified Mixture-of-Transformers architecture, supporting resolutions up to 720p across multiple aspect ratios for large-scale simulation, synthetic data generation, and policy learning.
https://aws.amazon.com/about-aws/whats-new/2026/01/cosmos3-edge-cosmos3-nano-cosmos3-super-on-sagemaker-jumpstart/
Related releases
- Meta's Muse-Glimmer-30B and Alibaba's Qwen 3.8-27B Models Now on SageMaker JumpStart AWS What's New ·
- SageMaker MLflow now supports customer managed keys AWS What's New ·
- SageMaker HyperPod Enhances Ray Support with Observability, Resilient Training, and Accelerated Inference AWS What's New ·
- AWS Glue 6.0 Reduces Pricing by 30% and Adds Iceberg v3 Support AWS What's New ·
- SageMaker AI Studio Adds Generative AI Inference Recommendations for Optimized Deployments AWS What's New ·
- AWS offers serverless solution to track SageMaker Studio costs with custom tags AWS Big Data Blog ·