NVIDIA Releases Nemotron 3.5 Super VL

NVIDIA has released Nemotron 3.5 Super VL, an open-weight multimodal AI model designed for specialized agents that work with text, images, documents, and video.The model combines visual understanding, reasoning, and tool use, allowing AI agents to process different types of information and complete complex, multistep tasks within a single workflow.

Architecture and Specifications

Nemotron 3.5 Super VL has 120 billion parameters, with 12 billion active parameters per token.It uses a hybrid architecture combining Mamba-2, Mixture-of-Experts (MoE), and Transformer attention layers, along with a C-RADIO visual encoder.The model supports:

• Input: text, images, and video.
• Output: text.
• Context window: up to 1 million tokens.
• Model formats: BF16 and NVFP4.
• Capabilities: reasoning, coding, tool use, and multimodal analysis.

The NVFP4 version uses quantization to reduce memory requirements, making the model more practical to deploy on a wider range of hardware.

What Nemotron 3.5 Super VL Can Do

The model is designed for specialized AI agents that need to process visual information, use external tools, and complete tasks involving multiple steps.Its primary use cases include:

• Document analysis: extracting information from PDFs, tables, charts, and scanned documents.
• Software development: analyzing code, identifying errors, and suggesting fixes.
• Web research: searching multiple sources, connecting information, and verifying answers.
• Video understanding: analyzing visual content for search, monitoring, and content review