GitHub radar
Nemotron 3.5 Super VL: one model for text, images, video, and code
Nvidia compressed its reasoning, coding, and document model so it needs less memory. It's one model for several tasks at once, but it only runs on data-center hardware, not a home PC.
The model reads text, images, and video. It uses that to write code, follow instructions, and work with documents.
When it helps
Useful if you're building an AI agent for code, documents, or video and you have access to Nvidia data-center GPUs (H100, H200) or a DGX Spark. Not for a regular PC or Mac — treat it as a preview of where agent models are heading.
Pros
- One model handles text, images, video, and code together — no need to combine separate tools
- NVFP4 compression cuts memory use compared to the full-size version
- Barely loses accuracy on some benchmarks, for example SWE-bench Verified
- Holds up to 1M tokens of context — text, images, and history at once
Cons
- Needs data-center hardware (H100, H200, or a DGX Spark) — a regular PC or Mac won't run it
- Results swing a lot between benchmarks: minor loss on some, real loss on others, for example WebVoyager
- No comparison with Opus or Fable — only Nvidia's own numbers, without independent verification
How to set it up — step by step
- 1Open your AI agent (Claude Code, Codex) in your project folder
- 2Send the agent the text from the block below
- 3Read the agent's answer about the model's code, document, and video skills
- 4If you don't have H100, H200, or DGX Spark hardware, treat this as information only — you can't install the model at home
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Open the NVIDIA Nemotron 3.5 Super VL model page (https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4) and explain what coding, document, and video tasks it is built for — it is a server-only model, so I cannot install it at home.
This is a server-only model — you cannot run it at home. You need one DGX Spark station, an H100 or H200 GPU, or a GB200 system.
Open on Hugging Face▌ More finds