Overview
Run NVIDIA Nemotron 3.5 Super VL 120B A12B with high-performance inference. Quickly deploy NVIDIA’s open multimodal model built for reasoning over text, images, documents, and video.
Explore additional models
Frequently
asked questions
Nemotron 3.5 Super VL 120B A12B is NVIDIA’s open multimodal model built for long-running agents, released in October 2026. It uses a hybrid Mamba-Transformer mixture-of-experts architecture with 120B total parameters and about 12B active per token. It reasons over text, images, video, documents, and GUI screenshots, with strong coding and tool-use skills for agentic work. On Crusoe it is available for high-performance Serverless Inference and Self-Serve Deployments with a 262,144-token context.
Choose Nemotron 3.5 Super VL 120B A12B for multimodal reasoning that needs depth, like complex documents, charts, and multi-step agents acting on screenshots. Nemotron 3 Super 120B A12B is the same size class but text-only, so it suits text-based agents. Nemotron 3 Nano Omni 30B A3B is smaller and faster, and also handles audio, which makes it a better fit for high-volume or audio-heavy media work.
Running Nemotron 3.5 Super VL 120B A12B with Serverless Inference is billed per token, with separate rates for uncached, cached, and output tokens, and no minimum commitment. Self-Serve Deployments are billed per GPU hour based on the GPU you choose. Images and video frames are converted to input tokens, so higher-resolution images and longer videos use more tokens. Access Nemotron 3.5 Super VL 120B A12B via Crusoe’s OpenAI-compatible API in Crusoe Intelligence Foundry.
Nemotron 3.5 Super VL 120B A12B brings Super-tier reasoning to visual inputs while activating only about 12B parameters per token, so it handles complex documents and video at a fraction of the compute of a dense model its size. NVIDIA built it for agentic workflows that combine visual understanding with coding and tool use.
Nemotron 3.5 Super VL 120B A12B is available for Serverless Inference and Self-Serve Deployments, no infrastructure setup required. It is not currently offered for Serverless Fine-Tuning. For an SLA-backed dedicated endpoint, talk with our team about Tailored Deployments.
The “VL” in Nemotron 3.5 Super VL 120B A12B stands for vision-language, meaning the model takes visual input alongside text and reasons across both. It pairs NVIDIA’s Super-tier language model with a vision encoder, so it can read documents, interpret charts and screenshots, and analyze video frames in the same request as a text prompt.


%204.png)
