NVIDIA Nemotron 3.5 Super VL Deployments Supported on Amazon EKS and SageMaker HyperPod EKS
NVIDIA Nemotron 3.5 Super VL, a new open multimodal model, is now deployable on Amazon EKS, EKS Auto Mode, and SageMaker HyperPod EKS. The model supports text, image, document, and video inputs with efficient compute scaling. Official recipes, weights, and deployment patterns are now available for customizable, self-hosted use.
What changed?
NVIDIA Nemotron 3.5 Super VL, an open 120B-parameter mixture-of-experts (MoE) multimodal model, is now validated and supported for deployment on Amazon EKS, EKS Auto Mode, and Amazon SageMaker HyperPod EKS. NVIDIA has released weights, training data, and post-training recipes, enabling direct customization or fine-tuning. The model supports large context windows up to one million tokens and accepts text, image, document, and video as input. Amazon and NVIDIA provide reference implementations using NVIDIA Dynamo, vLLM, and EFA acceleration for scalable inference.
Why does it matter to an everyday developer?
Developers can now self-host a large, open, multimodal model that supports complex, agentic AI workflows for tasks like code analysis, document intelligence, chart reading, and video understanding. Nemotron 3.5 Super VL is engineered for efficient compute, activating only 12 billion parameters per token while supporting long context use cases. Validated, step-by-step AWS blueprints and scaling patterns simplify integration into cloud-native applications or internal tools, without needing to rely on managed external APIs. However, hosting requires significant GPU resources and careful throughput planning.
What can the developer do now?
- Follow the official reference implementation to deploy Nemotron 3.5 Super VL on EKS, EKS Auto Mode, or SageMaker HyperPod EKS.
- Download weights, recipes, and documentation to experiment with custom fine-tuning for domain-specific agentic or multimodal applications.