Nemotron 3.5 Super VL arrives on SageMaker JumpStart with 1M-token context
Hi everyone, it's Shii-chan! A new vision-language model from NVIDIA has arrived on SageMaker JumpStart!
AWS What's New
aws.amazon.com
What was announced?
In AWS What's New, AWS announced that NVIDIA's Nemotron 3.5 Super VL is now available on Amazon SageMaker JumpStart. It's a vision-language model that accepts images, video, and text and generates text, so you can deploy multimodal AI on AWS.
Why it matters
The listed use cases are document intelligence, GUI understanding, video question answering, and multimodal agentic applications.
What changes
You can now deploy this large model from JumpStart. The two variants offered are the key choice!
- NVFP4: optimized inference on G7 instances with NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs
- BF16: reference weights on ml.p5en.48xlarge for customization such as SFT, RL, distillation, and domain adaptation
Dive Deep
Here is how the announcement describes the model:
- A 120B-parameter hybrid Mixture-of-Experts (MoE) language backbone with 12B active parameters, plus a vision encoder
- Context lengths of up to 1 million tokens
- A built-in Multi-Token Prediction (MTP) head for speculative decoding to speed up text generation
- Reasoning configurable per request
- Ten languages, including English, French, Spanish, German, Japanese, Hindi, Korean, and Chinese
- Post-trained with multimodal supervised fine-tuning on image, video, document, chart, and GUI data, followed by GRPO reinforcement learning
- Structured JSON output, tool calling, and configurable reasoning for enterprise agent workflows
Wrap-up
- Nemotron 3.5 Super VL is available on SageMaker JumpStart
- 120B (12B active) MoE with up to 1M tokens of context
- Two variants: NVFP4 for inference and BF16 for customization
This is for anyone building multimodal agents or document and GUI understanding on AWS!
