The 120B-parameter mixture-of-experts model activates 12B parameters per token and handles text, images, documents, and video.
Amazon Web Services now supports NVIDIA Nemotron 3.5 Super VL on Amazon EKS, EKS Auto Mode, and SageMaker HyperPod EKS, adding a 120B-parameter multimodal model to the Nemotron lineup already on the platform. The model combines reasoning and tool use with text, image, document, and video understanding.
NVIDIA distilled Nemotron 3.5 Super VL from Nemotron 3 Ultra, and the model joins Nemotron 3 Nano, Nano Omni, Super, Ultra, and Nemotron 3.5 Lightning. The hybrid Mamba-2, mixture-of-experts, and attention architecture pairs a RADIO vision encoder with layers that keep long contexts affordable, and NVIDIA publishes the weights, datasets, and recipes.
Sparse routing activates 12B of the 120B parameters per token, so compute per token matches a much smaller model even though GPU memory holds the full weights. A recursively shared multi-token prediction head supports native speculative decoding. Teams get agent workflows for coding, web research, document intelligence, cyber security, and video understanding on infrastructure they control.
NVIDIA reports 63.44 percent on Hermes Bench, 87.05 percent on LangChain agent evaluations, and 80.8 percent on CharXiv chart reasoning. Minimum GPU counts range from eight B200 or GB200 chips in BF16 down to one B300 or GB300 in NVFP4, and reference implementations connect the model to major open agent frameworks.
What matters
- NVIDIA Nemotron 3.5 Super VL now runs on Amazon EKS, EKS Auto Mode, and SageMaker HyperPod EKS.
- Sparse routing activates 12B of 120B parameters per token, so compute per token stays at small-model levels.
- NVIDIA publishes the weights, datasets, and post-training recipes for teams that want domain-specific tuning.
Why it matters
NVIDIA publishes the weights, datasets, and post-training recipes for teams that want domain-specific tuning.
This GenAI News article was prepared in original wording using reporting and materials published by AWS Machine Learning Blog. Source reference: https://aws.amazon.com/blogs/machine-learning/accelerate-agentic-ai-with-nvidia-nemotron-models-on-amazon-eks-and-sagemaker-hyperpod-eks/.
Drafted by the GenAI News review pipeline.