Nemotron 3.5 Super VL Ollama: How to Run It Locally

You can run Nemotron 3.5 Super VL with Ollama using one command — ollama run nemotron-3.5-super — and it supports text and images, tools and thinking, with up to a 1M-token context. The catch is size: it’s an 80–90GB model. My video below covers the model. Here’s how to run it with Ollama, what hardware you need and how to plug it into your agents.

Key takeaways

Command: ollama run nemotron-3.5-super.
Download size: about 80–90GB.
Supports vision, tools and thinking, up to 1M context.
Use it in Hermes or OpenCode via Ollama’s local endpoint.

How to run Nemotron 3.5 Super VL with Ollama

Install Ollama, then run:

ollama run nemotron-3.5-super

The library page lists the latest and 120b tags (the same model), tagged for vision, tools and thinking. Ollama’s API, Python and JavaScript examples use the same model name.

Hardware you need

Machine	Can it run Nemotron 3.5 Super VL?
Typical laptop (8–32GB)	No
High-memory Mac (96GB+ unified memory)	Possibly — test it
Workstation with large GPUs	Yes
Nvidia DGX Spark	Yes — Nvidia optimised a version for it

Usable context depends on your memory: prompts, images, history and tool results all share it.

Want this working in your business, not just bookmarked? Using free open models in your AI agents is exactly what we build together inside the AI Profit Boardroom — 3,400+ members, four live calls a week, and a full one-hour DeepSeek Harness course in the classroom.

Prefer it mapped 1-on-1 first? Book a free strategy session and we’ll plan it for your exact situation.

Join the AI Profit Boardroom →Claim Your Free SEO Strategy Session →

Connect Ollama to Hermes Agent

Ollama serves an OpenAI-compatible endpoint at http://localhost:11434/v1. In Hermes, add a custom endpoint with that address and the model name nemotron-3.5-super, ideally in its own profile. Hermes needs at least 64K of context for agent work, so set Ollama’s context size accordingly.

Why run it locally?

Privacy — client files never leave your machine.
No usage limits beyond your hardware.
No per-token cost once you have the hardware.
Offline work.

If you don’t have the hardware, use the free OpenRouter variant instead — see my free guide.

Ollama tips for big models

Close other heavy apps before loading it.
Start with a modest context window and increase it.
Use thinking mode only when a task needs it — it uses more tokens.
Keep a smaller local model for quick jobs.

Let me be clear: this is not a frontier model. It’s not going to replace Claude Opus 5.5 for your hardest work. What it is: a free, open, fast model that can read screenshots, PDFs and charts and then act on them — which is exactly the job a lot of agent work starts with.

Setting the context window

Ollama’s default context is much smaller than the model’s 1M-token maximum, and big contexts need lots of memory. Increase it step by step until you have enough for your task — agents typically need at least 64K — and watch your memory use. Remember that images, conversation history and tool results all count towards the context, so visual agent work fills it faster than plain chat.

Using images with Ollama

Because the model is tagged for vision, you can pass images through Ollama’s API or chat apps that support it. Send a screenshot, chart or document page along with your question. For multi-page PDFs, convert the pages to images and send the ones that matter. Keep images reasonably sized — they use context and memory.

When to choose Ollama over the cloud

Choose Ollama if privacy matters most — client documents, financial records, anything that shouldn’t leave your machine — or if you’ll run so much work that per-token costs add up. Choose the cloud if your hardware is limited or you only need it occasionally; the free OpenRouter variant is far easier. Many people do both: cloud for everyday work, local for sensitive files.

Final thoughts

Ollama also makes it easy to keep several models side by side. Pull Nemotron for visual and long-document work, keep a small, fast model for quick replies, and switch between them per task in your agent. That combination gives you a private, flexible local setup without relying on any cloud provider.

If you’re not sure your machine is big enough, check how much unified memory or GPU memory you have before downloading. As a rough rule, you want comfortably more free memory than the model’s download size, plus room for context and images.

Running a model this size at home is still new territory, so expect to tune settings the first time. Once it’s working, though, you have a private vision model with no usage limits — a genuinely useful tool for any business handling sensitive documents.

How I’d test it this week

If you have the hardware, pull the model tonight and test it on three private documents you’d never send to a cloud AI. That’s the use case local models are made for. If it’s too slow on your machine, keep it for overnight batch jobs and use the cloud version during the day.

Either way, connect it to Hermes as a separate profile so you can choose local or cloud per task.

Troubleshooting Ollama

If the download stalls, check your disk space — you need 80–90GB free plus room to spare. If loading fails or is very slow, your machine may not have enough memory; try a smaller context or use the cloud version. If images aren’t understood, make sure the app you’re using sends images through Ollama’s API rather than text. And if Hermes can’t connect, confirm Ollama is running and the endpoint is http://localhost:11434/v1 with the model name nemotron-3.5-super.

The bottom line on Nemotron 3.5 Super VL Ollama

Running Nemotron 3.5 Super VL on Ollama is one command if you have the memory. If you do, it’s a private, unlimited vision model for your agents.

FAQ: nemotron 3.5 super vl ollama

How do I run Nemotron 3.5 Super VL in Ollama?

Run ollama run nemotron-3.5-super.

How big is the Ollama download?

About 80–90GB.

Does the Ollama version support images?

Yes — it’s tagged for vision, tools and thinking.

Can I use it with Hermes Agent?

Yes — add Ollama’s local endpoint as a custom model in Hermes.

What if my computer is too small?

Use the free OpenRouter variant instead.

Is there a smaller Nemotron model for Ollama?

Nvidia’s Bolt 74B-A9B is designed to fit in 24GB of memory and supports Ollama; it’s a separate model from Super VL, but a good option for smaller machines.

Two ways I can help from here. Join the AI Profit Boardroom — it’s where your AI agent model setup gets built with 3,400+ members doing the same.

Or grab a free strategy session and bring your questions.