NVIDIA’s Nemotron 3.5 Super VL model now accessible on Amazon SageMaker JumpStart

NVIDIA’s Nemotron 3.5 Super VL model is now accessible through Amazon SageMaker JumpStart, broadening the array of foundational models available to AWS users. This advanced vision-language model from the Nemotron 3.5 series is capable of processing images, videos, and text inputs to generate text outputs, facilitating the deployment of multimodal AI solutions on AWS platforms.

The Nemotron 3.5 Super VL is specifically crafted for applications such as document intelligence, graphical user interface (GUI) comprehension, video question answering, and multimodal agentic tasks. It integrates a 120 billion-parameter hybrid Mixture-of-Experts (MoE) language backbone with 12 billion active parameters, alongside a vision encoder capable of handling context lengths up to one million tokens. Additionally, it features an integrated Multi-Token Prediction (MTP) head for speculative decoding, enhancing text generation speed with customizable reasoning per request. The model supports ten languages, including English, French, Spanish, German, Japanese, Hindi, Korean, and Chinese.

Post-training involves multimodal supervised fine-tuning on various data types such as images, videos, documents, charts, and GUIs, followed by GRPO reinforcement learning. This enables the model to provide structured JSON outputs, tool invocation, and customizable reasoning tailored for enterprise agent workflows. SageMaker JumpStart offers two model variants: the NVFP4, optimized for inference on G7 instances using NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, and the BF16 reference weights on ml.p5en.48xlarge, which are suitable for further customization through SFT, RL, distillation, and domain adaptation.

Through SageMaker JumpStart, users can effortlessly deploy the Nemotron 3.5 Super VL with minimal effort, addressing their unique AI requirements. To begin using these models, users can access the SageMaker JumpStart model catalog via the SageMaker console or utilize the SageMaker Python SDK to deploy them to their AWS account. For additional guidance on deploying and utilizing foundational models in SageMaker JumpStart, users are encouraged to refer to the Amazon SageMaker JumpStart documentation.