To run Nemotron 3.5 Super VL LM Studio style, open LM Studio, search for nvidia/nemotron-3.5-super-vl, hit Download, then load it and start chatting. The smallest build listed is Q4_K_M at about 70GB, so you need at least ~70GB of RAM plus extra for images and context. If your computer doesn't have that, stop here and use it hosted instead.
📺 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
LM Studio is the click-and-go option. No terminal, no commands. You get a proper chat window where you can drag in a screenshot or a PDF page and ask questions about it. That's the whole point of the VL version: it reads images.
If you want my full Agent OS setup with tutorials for swapping in open models like this, it's inside the AI Profit Boardroom ($69/mo).
What You Need Before You Start
Be honest with yourself about your machine first. This saves you a 70GB download that goes nowhere.
- Memory: at least ~70GB of RAM for the Q4_K_M build, and more on top for images and longer context
- Disk: about 70GB free for the file, plus headroom
- Machine type: a 128GB NVIDIA DGX Spark or a big-memory workstation is realistic. A typical laptop is not.
Why so big? Nemotron 3.5 Super VL has 120B total parameters. It's mixture-of-experts with 12B active per token, which makes it quicker, but every weight still has to sit in memory. Quantizing to Q4_K_M shrinks it to about 70GB. That's still huge.
Want the bigger picture on the model before you commit? Read my Nemotron 3.5 Super VL overview first.
How to Run Nemotron 3.5 Super VL LM Studio: Step by Step
Step 1: Download LM Studio
Go to the LM Studio website, download the app for your system and install it. Open it up.
Step 2: Search for the model
Click the search or discover area and type nvidia/nemotron-3.5-super-vl. That's the official listing name. Make sure you pick the NVIDIA one so you're not grabbing some random re-upload.
Step 3: Hit Get or Download
Choose the Q4_K_M build, the smallest one listed at about 70GB, and use the Download or Get button. Then wait. A 70GB file takes time even on fast internet. Go make a coffee. Maybe two.
Step 4: Load the model
Once the download finishes, load it from the model picker at the top of the chat window. LM Studio puts it into memory. If your machine doesn't have enough RAM, this is where it fails, and you'll know quickly.
Step 5: Attach an image and ask a question
Here's where the VL part earns its place. Attach a screenshot, a chart, a PDF page turned into an image or a photo of a receipt, then ask something specific:
- "What are the three biggest numbers in this chart?"
- "Turn this screenshot into an HTML page."
- "List every item and price on this receipt."
It reads the image and answers in text. NVIDIA built it to understand text, images, documents and video, with video handled through sampled frames or short clips.
📺 Watch: OpenClaw + Nvidia Nemotron 3 Super + Ollama is INSANE!
LM Studio vs Ollama for This Model
Both run it. They just suit different people.
| Feature | LM Studio | Ollama |
|---|---|---|
| Model name | nvidia/nemotron-3.5-super-vl | nemotron-3.5-super |
| Download size | Q4_K_M about 70GB | About 80-90GB |
| Memory needed | ~70GB+ RAM plus extra for images and context | Roughly 80-90GB plus extra |
| How you use it | Point-and-click app with a chat window | Terminal commands |
| Attaching images | Drag and drop in the chat | Through the terminal or a connected app |
| Agent launches | Not the focus | One-line launches for Hermes, OpenCode, Claude Code, OpenClaw |
| Best for | People who hate the terminal | People wiring it into agents |
My view: LM Studio if you want to chat with images and documents. Ollama if you want it running inside an agent. If agents are your thing, my Ollama setup for Nemotron 3.5 Super VL has the exact launch commands.
📺 Watch: New FREE Nvidia Nemotron 3 Super + Ollama
Tips to Keep It Running Smoothly
Context size uses memory
This is the big one. The model supports up to 1M tokens of context, but usable context depends on your memory and runtime. Every bit of context you set eats more RAM. If you crank context to the max on a 128GB machine, you can run out of memory even though the model itself loaded fine.
Start with a modest context length. Bump it up only when you actually need to feed it a long document.
Images use memory too
Every image you attach adds to the load. One screenshot is fine. Twenty high-resolution photos in one chat is asking for trouble. Start a fresh chat when you switch jobs.
Close the heavy stuff
Browsers with fifty tabs, video editors, other models running in the background. Close them. You want as much free memory as possible for the model.
Use thinking mode when it matters
The model has thinking and non-thinking modes. Thinking mode reasons more before answering, which helps on hard problems like chart reasoning. For quick questions, non-thinking is faster.
What It's Good At (and What It Isn't)
I've used Nemotron 3.5 Super VL free inside OpenCode, in my Agentic Operating System. I asked it to build a beautiful to-do list app, make it entertaining, engaging, and the build came out working.
My honest take: it's not a frontier model and it's not going to replace Opus 5.5. But it's free, open and fast, and it can read screenshots, PDFs, charts and receipts. So use it for jobs that start with a picture or a file.
NVIDIA reports 80.8% on CharXiv chart reasoning, and 84% with Python code interpretation. Those are NVIDIA's own numbers with no independent replication yet. For how I test models in the real world, see Goldie Bench.
Not sure where a model like this fits your business? Book a free AI strategy session and we'll work it out together.
Common Problems and Quick Fixes
- The model won't load: you don't have enough free RAM. Close other apps and try again. If it still fails, your machine is too small for this model. Go hosted.
- It loads, then crashes on a long chat: your context length is too high. Turn it down and start a fresh chat.
- Answers are very slow: the model is likely spilling out of fast memory. Shorter prompts and fewer images help. A bigger machine helps more.
- You can't find the listing: search the exact name nvidia/nemotron-3.5-super-vl and make sure LM Studio is up to date.
- The download stalls: check you have about 70GB of free disk space plus headroom, then restart the download.
If Your Machine Can't Run It
Most people's can't. That's fine. You've got easy options:
- OpenRouter: it's listed as nvidia/nemotron-3.5-super-vl, with a :free variant
- OpenCode free: pick "Nemotron 3.5 Super VL" in the model drop-down
- Hermes Agent: it shows up as one of the models you can pick
Same model, zero downloads, no 70GB RAM requirement. If you're thinking of buying hardware just for this, read my Nemotron 3.5 Super VL on DGX Spark breakdown first so you know what you'd actually be getting.
And if you want to build a full agent setup around free models like this, my Agent OS guide walks through running Hermes, OpenClaw, OpenCode and Claude side by side. The members inside the AI Profit Boardroom get the Agent OS zip, 4 weekly coaching calls and daily tutorials. Or grab a free AI strategy session if you'd rather talk it through.
FAQ
What is the model called in LM Studio?
It's listed as nvidia/nemotron-3.5-super-vl. Search for that exact name and pick the NVIDIA listing.
How much RAM do I need for Nemotron 3.5 Super VL in LM Studio?
At least ~70GB for the Q4_K_M build, plus extra for images and context. In practice that means a 128GB-class machine like a DGX Spark or a big-memory workstation.
Is there a smaller version I can run on a laptop?
Q4_K_M at about 70GB is the smallest build LM Studio names for this model. NVIDIA also released smaller Nemotron 3.5 Bolt edge models, but those are different models. For Super VL on a normal laptop, use OpenRouter or OpenCode free.
Can I send it PDFs?
It understands documents. The simplest way in LM Studio is to attach a page as an image, then ask your question about it.
