Yes, you can use Nemotron 3.5 Super VL free. The two easiest ways are picking it in the OpenCode model drop-down and using the :free variant on OpenRouter. The weights are also open, so you can download them at no cost, but running them yourself needs serious hardware. Below is every free route, what each one costs you in practice, and the catch with each.
📺 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've been using it free inside OpenCode and Hermes Agent as part of my Agentic Operating System. If you want that whole setup ready to go, it's inside the AI Profit Boardroom.
Every Way to Use Nemotron 3.5 Super VL Free
Here's the quick view before we go through each one.
| Free route | Cost | Catch |
|---|---|---|
| OpenCode model drop-down | Free | You can run out of free tokens; switch models when you do |
| OpenRouter :free variant | Free | Rate limits apply; check OpenRouter for current limits |
| Hermes Agent via OpenRouter | Free with the :free variant | Same OpenRouter rate limits |
| Open weights download | Free to download | Needs about 70-90GB of memory to run |
| Ollama or LM Studio locally | Free software | Same hardware bill; not a normal-laptop model |
1. Free in OpenCode
This is the fastest route and the one I used first. Open OpenCode, go to the model drop-down, and pick Nemotron 3.5 Super VL. That's it. No download, no hardware, no API key to sort out.
I asked it to build a beautiful to-do list app and make it entertaining and engaging. The build came out working. Because it's a vision-language model, you can also drop screenshots, PDFs, charts and receipts in and have it work from those. The full walkthrough is in my Nemotron 3.5 Super VL OpenCode guide.
2. OpenRouter's :free Variant
OpenRouter hosts the model as nvidia/nemotron-3.5-super-vl, and there's a :free variant alongside it. Use the free one by adding :free to the end of the model ID. You'll need an OpenRouter account and key, but you won't pay per token on the free variant.
The catch: free tiers come with rate limits. I'm not going to quote a number, because those limits are set by OpenRouter and can change. Check OpenRouter for what applies right now. For setup details and how the paid and free versions differ, read Nemotron 3.5 Super VL on OpenRouter.
3. Hermes Agent via OpenRouter
Once you've got OpenRouter set up, you can point Hermes Agent at the :free variant. It shows up as one of the models I can pick in Hermes, so I use it there too. This is handy when you've burned through your free OpenCode tokens and want to keep going.
If you run Ollama, there's also a launch command that drops the model straight into Hermes: ollama launch hermes --model nemotron-3.5-super. But that one runs locally, so the hardware catch below applies.
4. Download the Open Weights
NVIDIA released the model under the OpenMDW License Agreement v1.1 as open weights. The checkpoints are on Hugging Face (nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4) and ModelScope. NVIDIA also released the base, post-trained and NVFP4-quantized checkpoints, the training recipes, data and RL environments, 40 vision-language datasets on Hugging Face, and a cookbook on GitHub. All free.
The download is free. Running it isn't, unless you already own the hardware.
5. Run It Locally With Ollama or LM Studio
Both tools are free. In Ollama, the library name is nemotron-3.5-super, and you run it with ollama run nemotron-3.5-super. It's about 80-90GB. In LM Studio, search nvidia/nemotron-3.5-super-vl and hit Get; the smallest GGUF listed is Q4_K_M at about 70GB, and you need at least that much RAM plus extra for images and context.
Ollama can also launch it straight into other harnesses: ollama launch opencode --model nemotron-3.5-super, ollama launch claude --model nemotron-3.5-super, or ollama launch openclaw --model nemotron-3.5-super.
📺 Watch: New FREE Nvidia Nemotron 3 Super + Ollama
What "Free" Doesn't Cover
This is where people get caught out, so here it is straight.
- Hardware. The model is 120B parameters total with 12B active per token. Local runs need roughly 70-90GB of memory, so think a 128GB DGX Spark or a big-memory workstation. A normal laptop won't do it.
- Unlimited usage. The hosted free routes have limits. OpenCode can run out of free tokens. OpenRouter's :free variant has rate limits.
- Long context for nothing. The model supports up to 1M tokens of context, but how much you can actually use depends on your memory and runtime. Big contexts eat memory fast locally.
- Frontier-level results. It's free, fast and good at reading images and files. It's not a frontier model and it's not going to replace Opus 5.5 for your hardest work.
- Electricity and your time. Running a model this size locally isn't zero cost, even if the weights are.
For most people, the answer is simple: use it hosted through OpenCode or OpenRouter's free tier, and skip the local install unless you've already got the machine.
If you're trying to figure out which mix of free and paid models makes sense for your business, book a free AI strategy session.
📺 Watch: OpenClaw + Hunter Alpha & Healer Alpha FREE + NEW Nvidia Nemotron 3 Super + Ollama!
What to Do When You Hit the Free Limits
You will hit them at some point. Here's how I handle it.
- Switch to another free model in OpenCode. When I run out of tokens, I swap to another free model in the same drop-down. Big Pickle is one I use. Your work carries on.
- Move the job to Hermes. Point Hermes Agent at the OpenRouter :free variant and keep going there.
- Save Super VL for picture-and-file jobs. Don't spend your free allowance on plain text tasks another free model handles fine. Use it where it shines: screenshots, PDFs, charts, receipts and OCR.
- Go paid only if it's earning. If the free tier keeps running dry and the model is doing real work for you, the paid OpenRouter version or a cloud like Baseten, Fireworks AI or Together AI is the next step.
The trick is having several models wired in so you can swap without starting over. My Agent OS runs Hermes, OpenClaw, OpenCode and Claude side by side for exactly that reason. The full setup is in the Agent OS guide.
Is the Free Version Any Good?
Per NVIDIA, it scores 63.44% on Hermes Bench, ahead of the 56.33% for Nemotron 3 Ultra, and 80.8% on CharXiv chart reasoning. NVIDIA also reports up to 5x higher single-GPU throughput on agentic tasks than open models its size. Those are NVIDIA's own numbers with no independent replication of the full set yet, so test it on your own tasks. That's why I use Goldie Bench: same real jobs, every model, side by side.
The free route and the paid route run the same model, so the quality is the same. The difference is the limits. For the full picture on the model itself, read the main Nemotron 3.5 Super VL guide.
FAQ
Is Nemotron 3.5 Super VL really free?
Yes. You can use it free in OpenCode and through OpenRouter's :free variant, and the open weights are free to download. The free hosted routes have limits, and running it locally needs about 70-90GB of memory.
What's the easiest free way to try it?
OpenCode. Pick Nemotron 3.5 Super VL from the model drop-down and start prompting. No download or hardware needed.
What are the OpenRouter free limits?
Rate limits apply to the :free variant, and OpenRouter sets them. Check OpenRouter for the current limits rather than trusting a number from a blog post.
Can I run it free on my laptop?
The software and weights are free, but a normal laptop can't run it. The smallest local version is about 70GB. Use the hosted free routes instead.
Want the full Agent OS with every free model wired in, plus a 30-day roadmap, 4 weekly coaching calls and daily tutorials? Join 3,400+ members in the AI Profit Boardroom for $69/mo. Or talk it through on a free AI strategy session. Either way, Nemotron 3.5 Super VL free is worth trying today.
