Oriel — AI & technology worth noticing, every day.

NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart

Web page · https://aws.amazon.com/blogs/machine-learning/nvidia-nemotron-3-5-super-vl-is-now-available-in-amazon-sagemaker-jumpstart

NVIDIA Nemotron 3.5 Super VL, a 120B-parameter multimodal vision-language model, is now available in Amazon SageMaker JumpStart. Deploy the NVFP4 or BF16 variant in a few steps, then run text, image, and video inference for document intelligence, computer-use agents, and multimodal agentic workloads.

First seen by Oriel on .

AI summary

AWS has made NVIDIA Nemotron 3.5 Super VL, a multimodal model for document, GUI, video, and agentic workloads, deployable through Amazon SageMaker JumpStart.

  • The model combines a 120B-parameter hybrid MoE language backbone with 12B active parameters, a C-RADIO vision encoder, configurable reasoning, and a 1-million-token maximum context.
  • JumpStart offers a roughly 75 GB NVFP4 variant optimized for G7 inference and a roughly 249 GB BF16 variant intended for customization on ml.p5en.48xlarge.
  • NVFP4 matched BF16 on MMLongBench-Doc and remained within about one point on other highlighted multimodal benchmarks, sometimes scoring slightly higher.
  • With five concurrent requests, NVFP4 delivered 63.07–64.01 output tokens per second per request, while BF16 reached 140.85; mean first-token latency stayed near half a second.
  • JumpStart configurations provide 520–15,600-token contexts, accept one image and one video per request, require inline base64 media, and do not enable speculative decoding or LoRA adapters.

#amazon sagemaker #nemotron #vision-language models #model inference

Generated by gpt-5.6-sol. Check the original for details.

Aliases

SourceExternal IDURL
AWS Machine Learning Blog3e5b07e97d0a4f1f703ec7e89ab19f1b40aaa262https://aws.amazon.com/blogs/machine-learning/nvidia-nemotron-3-5-super-vl-is-now-available-in-amazon-sagemaker-jumpstart/ Read discussion