# What Is Nemotron 3.5? Nvidia’s Open Models for AI Agents

URL: https://whatisagenticai.net/blog/what-is-nemotron-3-5/
Published: 2026-10-19
Publisher: Little Tough LLC, maker of AI Agent Assistant: Agentic AI (https://apps.apple.com/us/app/ai-agent-assistant-agentic-ai/id6761501260)

Nemotron 3.5 is Nvidia’s family of free, open AI models built for agents. What Super VL, Bolt and Lightning are, what they need to run, and how to try them.

Nemotron 3.5 is the name on Nvidia’s newest batch of AI models, and it has been popping up in tech headlines since mid-October. Nvidia is best known for the chips that power AI. Now it is also giving away the models themselves.

So what is Nemotron 3.5, which version is which, and can a normal person use it? Here is the short, plain-English guide.

**TL;DR:** Nemotron 3.5 is Nvidia’s family of open AI models made for agents, meaning AI that uses tools, browses the web and finishes multistep tasks. The two newest members came out on October 15, 2026: **Nemotron 3.5 Super VL**, a 120 billion parameter model for company data centers, and **Nemotron 3.5 Bolt 74B-A9B**, a squeezed version that fits in 24GB of memory for local AI PCs. The weights are free to download from Hugging Face under the OpenMDW-1.1 license, which allows commercial use. You cannot run them on an iPhone.

## What is Nemotron 3.5?

Nemotron is Nvidia’s own line of AI models. Version 3.5 is a family, not one model. Nvidia says it built the newest models with the Nemotron Coalition, a group that includes Mistral AI, Thinking Machines Lab, Perplexity and Black Forest Labs.

The big idea is "open." Companies like OpenAI and Anthropic keep their best models on their own servers and charge for access. Nvidia publishes the Nemotron 3.5 files so anyone can download them, run them on their own machines and train them further on their own data. Nvidia’s Kari Briski told TechCrunch that companies can "post-train Super VL on their own data," which closed models do not allow.

## Nemotron 3.5 Super VL, Bolt and Lightning: which is which?

The names get confusing fast, so here is the whole family in one place.

| Model | Size | Made for | Released |
| --- | --- | --- | --- |
| Nemotron 3.5 Super VL | 120B total, 12B active | Enterprise agents. Reads text, images and video up to 2 minutes. Up to 1M token context. | October 15, 2026 |
| Nemotron 3.5 Bolt 74B-A9B 24GB | 74B total, 9B active | Local agents on AI PCs: tool calling, web browsing, terminal use. Up to 256K tokens. | October 15, 2026 |
| Nemotron 3.5 Bolt E4B 4GB | Fits in 4GB | Fast voice assistants and interactive characters | Nvidia says later in October |
| Nemotron 3.5 Lightning 30B-A3B | 30B total, 3B active | A smaller, fast model that runs on one big GPU | August 11, 2026 |

"Active" is the key word. These models use a design called mixture of experts. The model holds many small specialist parts, but only a few switch on for each word it writes. Super VL has 120 billion parameters in total, yet only about 12 billion do work at any moment. That makes it faster and cheaper to run than its size suggests.

Bolt is the clever one. Nvidia’s model card says it was compressed from Super VL and stored with 2-bit weights so the whole thing fits in 24GB. That is small enough for the new AI PCs built on Nvidia’s RTX Spark chip and for DGX Spark desktop workstations.

## Why Nvidia says it is built for agents

Nvidia did not pitch Nemotron 3.5 as a chatbot. Every model card talks about agent work: calling tools, browsing, using a terminal and reading long documents. Nvidia says Super VL finishes some tasks faster than models from Anthropic and OpenAI once it has been post-trained for a specific job, such as document processing, financial accounting or medical insurance claims. It also named early users including EY, Ramp, CrowdStrike, Perplexity, Salesforce and Skydance. Those are Nvidia’s own claims.

Independent numbers give a calmer picture. According to The Decoder, Super VL scores 26 on the Artificial Analysis Intelligence Index and Bolt scores 21, while Claude Opus 5.5 leads that index at 58. So Nemotron 3.5 is not trying to be the smartest model in the world. It is trying to be a strong, free model that a company can own and tune.

## Is Nemotron 3.5 free?

The models are free to download. Nvidia publishes them on Hugging Face under the OpenMDW License Agreement, version 1.1, and each model card says the model "is ready for commercial or non-commercial use." Read the license yourself before you build a product on it.

Free does not mean cheap to run, though. Nvidia’s card for Super VL lists 4 H100 GPUs or 2 GB200 GPUs for deployment. Those are data center chips. Bolt is far lighter, but it still wants 24GB of memory and a Linux machine with a recent Nvidia GPU.

## How to try Nemotron 3.5

- **Download it:** search for "NVIDIA-Nemotron-3.5" on Hugging Face. Super VL comes in BF16 and NVFP4 versions. Bolt also has a GGUF version for local tools.

- **Run it locally:** Bolt’s model card lists vLLM on Nvidia A100, H100 or Blackwell hardware. The smaller Lightning model runs on a single 80GB H100 or A100, and its card links to community versions for llama.cpp, Ollama and LM Studio.

- **On a phone:** not today. Even the 4GB Bolt E4B is aimed at RTX cards, Jetson modules and Spark machines, not iPhones.

## Nemotron 3.5 vs Muse Spark Open

The closest rival this month is Meta’s [Muse Spark Open](https://whatisagenticai.net/blog/what-is-muse-spark-open/), which also came out on October 15.

|  | Nemotron 3.5 Super VL | Muse Spark Open |
| --- | --- | --- |
| Maker | Nvidia | Meta |
| Size | 120B total, 12B active | 816B total, about 42B active |
| Context | Up to 1M tokens | 1M tokens |
| License | OpenMDW-1.1, commercial use allowed | Meta’s Muse Model License |
| Small local version | Yes, Bolt (24GB) and Bolt E4B (4GB) | No |

The short version: Meta went big, and Nvidia went small and efficient, with a version made for laptops.

## What this means for you

If you are not a developer, you will probably never download Nemotron 3.5. But you will feel it. When a chip giant gives away agent models for free, it gets cheaper for every company to build AI that does real tasks. That is the core of [agentic AI](https://whatisagenticai.net/what-is-agentic-ai/): AI that plans steps and finishes the job instead of only chatting.

It is also a reminder of how fast the list moves. Today alone brought [MiniMax M3.1 Flash](https://whatisagenticai.net/blog/what-is-minimax-m3-1-flash/), and last week brought Muse Spark Open and a DeepSeek update. You do not need a GPU or a spreadsheet of model names to get value from agents.

Our app, **Agentic AI**, puts a team of ready-made expert agents on your iPhone and iPad, like a Research Analyst, a Writing Coach and a Code Helper. Give a task to one agent or put up to 4 on it together. There is no account to make, and you get 3 free tasks to try it.

## Quick answers

**What is Nemotron 3.5?** A family of open AI models from Nvidia, built for AI agents. It includes Super VL, Bolt and Lightning.

**When was Nemotron 3.5 Super VL released?** Nvidia released Super VL and Bolt 74B-A9B on October 15, 2026. Nemotron 3.5 Lightning came out earlier, on August 11, 2026.

**Is Nemotron 3.5 Super VL free?** Yes, the weights are free on Hugging Face under the OpenMDW-1.1 license, which allows commercial use. Running it takes powerful data center GPUs.

**What is Nemotron 3.5 Bolt?** A compressed version of Super VL with 74 billion parameters, 9 billion active, that fits in 24GB of memory so it can run agents on local AI PCs.

**Can I use Nemotron 3.5 on my iPhone?** No. Nvidia aims these models at its own GPUs and Spark machines. For agents on iPhone, use an [agent app](https://whatisagenticai.net/blog/ai-agents-on-iphone/) instead.

**Sources:** [Nvidia model card: Super VL](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16) · [Nvidia model card: Bolt 74B-A9B](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB) · [Nvidia model card: Lightning](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16) · [TechCrunch](https://techcrunch.com/2026/10/19/nvidia-launches-new-open-models-to-run-agents-in-the-enterprise-and-on-pcs/) · [The Decoder](https://the-decoder.com/nvidias-nemotron-3-5-super-vl-is-an-open-model-built-to-run-enterprise-ai-agents-nonstop/) · [TechRepublic](https://www.techrepublic.com/article/news-nvidia-nemotron-3-5-super-vl-bolt-ai-models/)