NVIDIA Nemotron 3.5 Super VL Available in Amazon SageMaker JumpStart
Frontier AI models · Enterprise AI
Amazon SageMaker JumpStart now provides NVIDIA Nemotron 3.5 Super VL, a hybrid Mixture-of-Experts vision-language model. Users can deploy an NVFP4 quantized checkpoint for optimized inference or full-precision BF16 reference weights.
AWS announced that NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart. The vision-language model pairs a hybrid Mixture-of-Experts (MoE) language backbone with a C-RADIO vision encoder. It accepts image, video, and text inputs to generate text, supporting context lengths of up to 1 million tokens and a built-in Multi-Token Prediction (MTP) head. JumpStart provides two variants: an NVFP4 quantized version optimized for inference on G7 instances, and full-precision BF16 reference weights for customization on P5en instances. NVIDIA designed the model for document intelligence, graphical user interface (GUI) understanding for computer-use agents, video question answering, and multimodal agent workflows.
Key takeaways
- Nemotron 3.5 Super VL combines a hybrid MoE language backbone with a C-RADIO vision encoder to accept text, image, and video inputs.
- SageMaker JumpStart offers two variants: NVFP4 for optimized inference on G7 instances, and BF16 reference weights on P5en instances.
- The model supports context lengths up to 1 million tokens, targeting document intelligence, GUI understanding, and multimodal agent tasks.
- Users can deploy via SageMaker Studio or the SageMaker Python SDK, with configurable reasoning capabilities per request.
Source：AWS Machine Learning Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human.