Yes, you can run Nemotron 3.5 Super VL Hermes Agent setups today, and the easiest way is through OpenRouter: run hermes model, pick OpenRouter, paste your key, and choose nvidia/nemotron-3.5-super-vl or the free variant. If you have a big-memory machine, you can run it locally through Ollama instead. This page shows you all three routes and what jobs to actually give it once it is running.
📺 Watch: NVIDIA Nemotron 3.5 Super VL Is INSANE! (FREE AI Model)
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside
I use this model inside Hermes myself. It shows up as one of the models I can pick, right next to the others in my stack. It is not a frontier model. It is not going to replace Opus 5.5. But it is free, it is open, it is fast, and it can read screenshots, PDFs and charts. That last part is why it earns a slot in Hermes.
If you want the full Agent OS I run Hermes inside, with the zip, the video walkthrough and a 30-day roadmap, it lives inside the AI Profit Boardroom ($69/mo). More on that further down.
What Nemotron 3.5 Super VL brings to Hermes
Quick background so you know what you are plugging in. Nemotron 3.5 Super VL is NVIDIA's open multimodal model built for agents. It has 120B total parameters in a mixture-of-experts design, with 12B active per token. It reads text, images, documents and video frames, and it writes text back. Reasoning, instruction following and tool use all live in one model, with thinking and non-thinking modes.
For a Hermes user, the number that matters most is this: per NVIDIA, it scores 63.44% on Hermes Bench, up from 56.33% for Nemotron 3 Ultra. That is NVIDIA's own figure, and nobody has independently replicated it yet, so treat it as a strong signal, not a verdict. If you want the full model breakdown first, read my Nemotron 3.5 Super VL overview.
Route 1: Nemotron 3.5 Super VL Hermes Agent setup through OpenRouter
This is the route I recommend for almost everyone. No downloads, no 80GB of memory, no drivers. Hermes talks to OpenRouter by default, and OpenRouter hosts the model.
- Get an API key from openrouter.ai.
- In your terminal, run hermes model. This is the command Hermes uses to choose your provider and model.
- Pick OpenRouter from the provider list and paste your API key when asked.
- Choose the model. Use nvidia/nemotron-3.5-super-vl for the paid version, or nvidia/nemotron-3.5-super-vl:free for the free tier.
- Start Hermes and send a test message to confirm it responds.
Prefer setting the key directly? Hermes also lets you run hermes config set OPENROUTER_API_KEY followed by your key. It saves secrets to the right file automatically.
Already inside a Hermes session? Type /model. That opens the model picker so you can switch mid-session without leaving the chat. I do this all the time when a job needs a different brain.
One thing to know about the free variant: free tiers usually come with tighter rate limits. Check the OpenRouter model page for current limits and pricing before you build anything that runs all day.
📺 Watch: Claude + Hermes Agent = Superpowers!
Route 2: Run it locally with Ollama
This route is only for people with serious hardware. The Ollama library name is nemotron-3.5-super, and it is the same vision model. The download is roughly 80-90GB, so you need a machine with that much memory to spare, like a 128GB DGX Spark or a big-memory workstation. A normal laptop will not run it.
If you have the hardware, Ollama makes the Hermes hookup one line:
ollama launch hermes --model nemotron-3.5-super
That launches Hermes wired to the local model. You get privacy, no per-token bill and no rate limits. The trade-off is speed on anything less than top-end hardware. I cover the full local install, sizes and tags in my Nemotron 3.5 Super VL Ollama guide.
📺 Watch: Hermes NEW AI Super Agent is INSANE! 🤯
Route 3: Add it to Agent OS as its own profile
My Agentic Operating System runs Hermes, OpenClaw, OpenCode and Claude side by side. When a new open model drops, I do not rip out my setup. I just add it.
The clean way is a dedicated Hermes profile. Hermes supports multiple profiles, each with its own config. Run hermes profile create vision (or any name you like), then run hermes -p vision setup (the -p flag works with any Hermes command) and point that profile at Nemotron 3.5 Super VL through OpenRouter. Now you have a vision agent that does not touch your main setup.
If you would rather not touch the terminal, just ask your AI agent inside Agent OS to do it. Paste something like this:
Create a new Hermes profile called vision that uses OpenRouter with the model nvidia/nemotron-3.5-super-vl:free. Keep my main profile exactly as it is. Then test it with a one-line message and tell me if it works.
That is how I want people setting this up. Let the agent do the config. The full walkthrough of how the workspace fits together is in my Agent OS guide.
What jobs to give it in Hermes
Here is the simple rule: give it the jobs that start with a picture or a file. That is where this model pulls its weight.
- Screenshots. Drop in a screenshot of a dashboard, an error, or a competitor's landing page and ask Hermes to pull out what matters. Hermes lets you paste images from the clipboard in supported terminals.
- PDFs and documents. Contracts, reports, invoices, slide decks. Ask for a summary, a list of action items, or the three numbers that matter.
- Charts. Per NVIDIA, it scores 80.8% on CharXiv chart reasoning, and 84% with Python code interpretation. Hand it a traffic chart and ask what changed and when.
- Agent tasks with tools. It was built for tool use. NVIDIA reports 87.05% on LangChain evals. Use it for research runs, file sorting and multi-step jobs that do not need a frontier model.
What I would not give it: your hardest reasoning or coding work where one mistake costs you money. Keep a frontier model for that. You can see how models stack up on my own tests in Goldie Bench.
If you want help mapping which of your daily jobs should go to which model, book a free AI strategy session and we will work it out for your business.
Tips that save you headaches
Keep one profile per model
Do not keep swapping the model on your main profile. Make a separate profile for Nemotron. Your main agent stays stable, and you can compare outputs side by side.
Set a fallback model
Free tiers run out. Endpoints get busy. Hermes supports fallback providers in its config, so if Nemotron is unavailable, the job rolls over to another model instead of dying. Ask your agent to add a fallback for you.
Use the thinking mode wisely
The model has thinking and non-thinking modes. Thinking helps with messy documents and multi-step plans. For quick extraction from a clean screenshot, non-thinking is faster.
Test before you trust
Every benchmark here comes from NVIDIA. Run it on five of your real tasks before you move anything important onto it. That tells you more than any leaderboard.
Where it fits against other Hermes models
Nemotron 3.5 Super VL is a strong free option with vision built in. That is its edge. Plenty of models in Hermes write well. Fewer can read a screenshot and act on it in the same run, for free.
But it is one option in a stack, not the whole stack. If you are picking the main brain for your Hermes setup, compare it against the others in my best Hermes Agent models rundown. My setup uses a frontier model for the hard stuff and free open models like this one for the volume work.
Want the exact Agent OS I use, with Hermes, OpenCode, OpenClaw and Claude already wired up? It comes inside the AI Profit Boardroom: the Agent OS zip, a video tutorial, a 30-day roadmap, updates as new models drop, four weekly coaching calls and daily tutorials, with 3,400+ members already inside.
FAQ
How do I add Nemotron 3.5 Super VL to Hermes Agent?
Run hermes model, pick OpenRouter, paste your API key, and choose nvidia/nemotron-3.5-super-vl or nvidia/nemotron-3.5-super-vl:free. Inside a running session, type /model to switch to it.
Can I run Nemotron 3.5 Super VL in Hermes for free?
Yes. Use the :free variant on OpenRouter, or run it locally through Ollama if you have roughly 80-90GB of memory. Check the OpenRouter page for current free-tier limits.
What is the Ollama command for Hermes?
ollama launch hermes --model nemotron-3.5-super. In Ollama the model is listed as nemotron-3.5-super, and it is the same vision model.
Is Nemotron 3.5 Super VL good enough to be my main Hermes model?
For vision and document jobs, it is a strong free pick. For your hardest reasoning and coding, keep a frontier model as your main brain and use Nemotron as a dedicated profile. If you want a second opinion on your setup, grab a free AI strategy session.
That is the full playbook for a Nemotron 3.5 Super VL Hermes Agent setup: OpenRouter for most people, Ollama for big machines, and its own profile inside Agent OS so the rest of your stack stays untouched.
