Nemotron 3.5 Super is the name Ollama uses for NVIDIA's new vision-language model, which NVIDIA itself calls Nemotron 3.5 Super VL. Same model, two names. It has 120B total parameters, 12B active per token, takes text and images in, and handles up to 1M tokens of context.
📺 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
If you searched "Nemotron 3.5 Super" and got confused by all the VL, Bolt, Nano, Omni and Ultra names flying around, this page clears it up. I will show you which name means what, how the family fits together, and which one you should actually pick.
Nemotron 3.5 Super vs Super VL: Same Model
Here is the short version. NVIDIA released Nemotron 3.5 Super VL on 15 October 2026. The "VL" means vision-language. When Ollama added it to its library, it listed it simply as nemotron-3.5-super.
Ollama describes it as "NVIDIA's vision-language model for coding, reasoning, tool use, and document understanding." The tags on the listing are vision, tools and thinking. Size is 120b, about 80-90GB to download, with a 1M context window and text plus image input.
So if you see "Nemotron 3.5 Super" in Ollama and "Nemotron 3.5 Super VL" on OpenRouter, LM Studio or Hugging Face, you are looking at the same vision model. You are not missing a separate text-only version.
Here is how the name shows up in each place:
- Ollama: nemotron-3.5-super
- OpenRouter: nvidia/nemotron-3.5-super-vl (plus a :free variant)
- LM Studio: nvidia/nemotron-3.5-super-vl
- Hugging Face: nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4
For the full breakdown of what the model does, read the main Nemotron 3.5 Super VL guide. And if you want to learn how to put models like this to work in real agent setups, that is what we do in the AI Profit Boardroom.
Why the Names Get Confusing
NVIDIA now has two generations of Nemotron models live at the same time, and each one has several sizes. On top of that, every platform shortens names in its own way. Add "Super" appearing in both generations and it is easy to grab the wrong one.
The fix is simple. Look at the version number first, then the tier.
- 3.5 is the new generation, released 15 October 2026.
- 3 is the older generation.
- Super, Ultra, Nano, Bolt are the tiers inside each generation.
So "Nemotron 3 Super" is the older model. "Nemotron 3.5 Super" is the new vision model. NVIDIA reports the new one beats Nemotron 3 Super on SWE-bench and Terminal-Bench setups, though it gave no exact numbers for those.
📺 Watch: OpenClaw + Nvidia Nemotron 3 Super + Ollama is INSANE!
The Nemotron 3.5 Family
Alongside Super VL, NVIDIA released smaller Nemotron 3.5 Bolt edge models. These are built for much smaller hardware.
- Nemotron 3.5 Super VL is the flagship of this release: 120B total, 12B active, vision plus agents, distilled from Nemotron 3 Ultra.
- Nemotron 3.5 Bolt 74B-A9B is aimed at 24GB systems.
- Nemotron 3.5 Bolt E4B is aimed at 4GB edge hardware.
The Bolt models are there so you can run Nemotron 3.5 on devices that could never hold the 120B Super. If you only have a normal machine, those are the ones to watch.
📺 Watch: New FREE Nvidia Nemotron 3 Super + Ollama
The Older Nemotron 3 Models
The Nemotron 3 generation is still around and you will see it referenced in NVIDIA's own comparisons.
- Nemotron 3 Ultra is the big one. Super VL was distilled from it. Per NVIDIA, the new Super VL actually beats Ultra on Hermes Bench (63.44% vs 56.33%) and LangChain evals (87.05% vs 84.3%).
- Nemotron 3 Super is the older Super. Super VL beats it on coding and terminal setups, NVIDIA says.
- Nemotron 3 Nano Omni is the small multimodal model. On CharXiv chart reasoning, Super VL scores 80.8% vs 63% for Nano Omni, per NVIDIA.
All of those numbers are NVIDIA's. Nobody has independently replicated them yet. I run my own practical tests on Goldie Bench if you want a second opinion on business tasks.
Which Nemotron Should You Pick?
Here is the whole family side by side, using only what NVIDIA and the listings have stated.
| Model | Size | Best for |
|---|---|---|
| Nemotron 3.5 Super VL (Ollama: Nemotron 3.5 Super) | 120B total, 12B active | Agents that read screenshots, PDFs, charts and video; coding and tool use |
| Nemotron 3.5 Bolt 74B-A9B | 74B total, 9B active | Running Nemotron 3.5 on 24GB systems |
| Nemotron 3.5 Bolt E4B | E4B | Edge devices with 4GB of memory |
| Nemotron 3 Ultra | Largest of Nemotron 3 | The older flagship; the teacher Super VL was distilled from |
| Nemotron 3 Super | Older Super tier | Existing setups already built on it |
| Nemotron 3 Nano Omni | Nano tier | Older small multimodal option |
My simple rule:
- If you want the best new open Nemotron for agents and have a hosted option or a big machine, pick Nemotron 3.5 Super (Super VL).
- If you only have a 24GB GPU or an edge device, look at the Bolt models.
- If you are starting fresh, skip the Nemotron 3 models. The 3.5 Super beats them on NVIDIA's own agent and chart tests.
Running Nemotron 3.5 Super in Ollama
If you have the hardware, Ollama is the easiest way to run it locally. You need around 80-90GB of memory, so think a 128GB DGX Spark or a big-memory workstation. A normal laptop will not cope.
The basic command is: ollama run nemotron-3.5-super
Ollama can also launch it straight into an agent harness:
- ollama launch hermes --model nemotron-3.5-super
- ollama launch opencode --model nemotron-3.5-super
- ollama launch claude --model nemotron-3.5-super
- ollama launch openclaw --model nemotron-3.5-super
The full setup is in the Nemotron 3.5 Super VL Ollama guide.
No big machine? Use it hosted instead. OpenRouter has a free variant, and OpenCode lets you pick Nemotron 3.5 Super VL from the model drop-down for free. That is how I tested it.
What I Found Testing It
I picked it in OpenCode inside my Agentic Operating System and asked it to "build a beautiful to-do list app, make it entertaining, engaging." It built the app and it worked. I also use it inside Hermes Agent, where it shows up in the model list.
My honest take: it is not a frontier model and it is not going to replace Opus 5.5. But it is free, open, fast, and it can read screenshots, PDFs and charts. Use it for jobs that start with a picture or a file. If you run out of free tokens in OpenCode, switch to another free model there like Big Pickle, or run it in Hermes.
My Agent OS runs Hermes, OpenClaw, OpenCode and Claude side by side, so swapping in a new model like this takes seconds. The Agent OS guide shows how the pieces connect.
Not sure which model belongs in which part of your business? Book a free AI strategy session and we will sort it out.
If you want the full Agent OS zip, a video tutorial, a 30-day roadmap, updates, 4 weekly coaching calls and daily tutorials, join the AI Profit Boardroom ($69/mo) with 3,400+ members.
For the business side of the launch, read the NVIDIA Nemotron 3.5 Super VL release breakdown.
FAQ
Is Nemotron 3.5 Super the same as Nemotron 3.5 Super VL?
Yes. Ollama lists NVIDIA's Nemotron 3.5 Super VL vision-language model under the name nemotron-3.5-super. It has the vision, tools and thinking tags and takes text and image input.
Is there a text-only Nemotron 3.5 Super?
Not in the listings covered here. The model Ollama calls Nemotron 3.5 Super is the vision-language model. The older text model you may be thinking of is Nemotron 3 Super.
Which Nemotron runs on a normal computer?
The Nemotron 3.5 Bolt models. NVIDIA built the 74B-A9B for 24GB systems and the E4B for 4GB edge hardware. The 120B Super needs around 70-90GB of memory.
Is Nemotron 3.5 Super better than Nemotron 3 Ultra?
On agent benchmarks, NVIDIA says yes: 63.44% vs 56.33% on Hermes Bench and 87.05% vs 84.3% on LangChain evals. Those are NVIDIA's own numbers with no independent replication yet.
Want a clear plan for using free open models in your agents? Grab a free AI strategy session. Whatever name you see it under, Nemotron 3.5 Super is worth a slot in your stack.
