NVIDIA Nemotron 3.5 Super VL Available in Amazon SageMaker JumpStart

The landscape of enterprise AI shifted once more with the official arrival of NVIDIA Nemotron 3.5 Super VL in Amazon SageMaker JumpStart. As a powerhouse 120B-parameter multimodal vision-language model, Nemotron 3.5 Super VL brings advanced document intelligence, computer-use agents, and highly complex multimodal reasoning directly to AWS enterprise environments. For developers and systems architects striving to build next-generation AI workflows, this release offers an exciting blueprint for deploying heavyweight models in production. However, managing infrastructure for a 120B model via dedicated cloud instances can introduce heavy financial overhead and operational complexity. This is where API aggregation platforms like n1n.ai bridge the gap, providing immediate access to cutting-edge models without the burden of cluster orchestration.

Understanding NVIDIA Nemotron 3.5 Super VL

NVIDIA has positioned Nemotron 3.5 Super VL at the absolute bleeding edge of open-weight vision-language modeling. With 120 billion parameters, the model is architected specifically to process heavy interleaved streams of text, high-resolution images, and video sequences. Unlike smaller models that struggle with fine-grained spatial grounding or dense document parsing, Nemotron 3.5 Super VL excels in scenarios requiring structural layout analysis, chart comprehension, and interactive GUI navigation for computer-use agents.

Key capabilities unlocked by this release include:

• Advanced Document Intelligence: Parsing complex multi-column PDFs, financial balance sheets, and handwritten annotations with high fidelity.
• Computer-Use Agent Workloads: Translating visual desktop elements into actionable API calls or UI click coordinates for automation workflows.
• Multimodal Agentic Reasoning: Correlating video timelines with spoken prompts to execute long-horizon enterprise tasks.

When deploying within Amazon SageMaker JumpStart, engineers can choose between two primary precision formats: the memory-efficient NVFP4 variant or the full-precision BF16 variant, depending on throughput and latency budgets.

Deployment Walkthrough in Amazon SageMaker JumpStart

Deploying Nemotron 3.5 Super VL through SageMaker JumpStart requires proper GPU provisioning due to the model's 120B size. Typically, you will need multi-GPU instances such as ml.p5.48xlarge nodes featuring NVIDIA H100 GPUs to achieve acceptable inference latency.

Step 1: Accessing the JumpStart Console

Navigate to the Amazon SageMaker console in your preferred AWS region, select JumpStart from the left-hand navigation pane, and search for Nemotron-3.5-Super-VL.

Step 2: Selecting the Quantization Variant

You will be presented with configuration options for deployment. Choose between BF16 for uncompromising precision or NVFP4 for optimized memory footprints.

import sagemaker
from sagemaker.jumpstart.model import JumpStartModel

role = sagemaker.get_execution_role()
model_id = "nvidia-nemotron-3-5-super-vl-nvfp4"

# Initialize the JumpStart model instance
model = JumpStartModel(model_id=model_id, role=role)

# Deploy to a dedicated endpoint with H100 hardware
predictor = model.deploy(
    initial_instance_count=1,
    instance_type="ml.p5.48xlarge