Nvidia Nemotron 3.5 Super VL Now Available on Amazon SageMaker JumpStart

At a glance

Nvidia's AI model Nemotron 3.5 Super VL, which understands images, video, and text together, can now be deployed directly on Amazon SageMaker JumpStart. The model takes images, video, and text as input and responds with text.

Key points

1. 1

   The model has 120 billion parameters in total but uses only 12 billion at a time. It selects only the expert modules it needs for each calculation.

2. 2

   It is built for tasks such as reading and analyzing documents, agents that understand buttons on a screen, and answering questions about videos.

3. 3

   On JumpStart, you can choose between an NVFP4 version tuned for fast responses and a BF16 version used for fine-tuning.

4. 4

   The length of text it can process at once expands to up to 1 million tokens (small text units), and you can decide per request whether to use reasoning (where the AI works through steps before answering).

Why it matters

This gives developers and companies building tools that handle photos, videos, and documents more practical options.