Nemotron 3.5 Super VL is Nvidia’s new free, open AI model that can read your screenshots, PDFs and charts — and then actually do the work from there. On Hermes Bench, a test built for Hermes Agent, it beat Nvidia’s own bigger model, Nemotron 3 Ultra, and you can use it free in OpenCode and Hermes today. I show it in the video below. Here’s what it is, the numbers that matter, and where it fits in your agent stack.
Key takeaways
- 120B parameters, 12B active — big-model brains, small-model speed.
- Reads text, images, documents and short videos in one model.
- 63.44% on Hermes Bench vs 56.33% for Nemotron 3 Ultra (Nvidia-reported).
- Free in OpenCode and via OpenRouter’s free variant; open weights on Hugging Face.
What Nemotron 3.5 Super VL is
Nvidia released it on 15 October 2026 as part of its Nemotron family of open models. “VL” means vision-language: it understands images and documents as well as text. Nvidia says it’s comparable with its largest reasoning model at roughly a fifth of the parameter count.
| Nemotron 3.5 Super VL at a glance | |
|---|---|
| Made by | Nvidia |
| Released | 15 October 2026 |
| Size | 120B total parameters, 12B active per token (mixture of experts) |
| Architecture | Hybrid Mamba-2 + Transformer MoE with a C-RADIO vision encoder and multi-token prediction |
| Context | Up to 1M tokens |
| Inputs | Text, images, document pages and MP4 video up to 2 minutes (text output) |
| Reasoning | Off, medium or on; supports tool calling |
| Licence | OpenMDW-1.1 — open weights, commercial use allowed |
| Where to get it | Hugging Face, OpenRouter (incl. a free variant), Ollama, LM Studio, NVIDIA NIM, ModelScope |
Why a vision model matters for agents
Most agents run on text-only models. Hand a text model a screenshot, a chart or a scanned receipt and it’s stuck. A lot of real agent work starts with something visual: a client report, a dashboard screenshot, a PDF. Nemotron 3.5 Super VL puts text and visual understanding in one model, so your agent doesn’t have to pass context between a separate text model and vision model.
On CharXiv — a test of whether an AI really understands charts — it scores 80.82, and it’s strong at reading text inside images (OCR), like the text on a receipt photo.
Want this working in your business, not just bookmarked? Using free open models in your AI agents is exactly what we build together inside the AI Profit Boardroom — 3,400+ members, four live calls a week, and a full one-hour DeepSeek Harness course in the classroom.
Prefer it mapped 1-on-1 first? Book a free strategy session and we’ll plan it for your exact situation.
Join the AI Profit Boardroom →Claim Your Free SEO Strategy Session →
The three numbers to know
- 120 billion parameters in total, but only 12 billion switch on for each token. Picture a team of 120 where only the 12 best-suited people stand up for each job — that’s why it’s fast.
- Up to 5× faster — Nvidia says it’s five times faster than the next-best open agentic model in its class, helped by multi-token prediction, where the model drafts several words ahead and checks them in one go.
- 63.44% on Hermes Bench, versus 56.33% for Nemotron 3 Ultra.
Nemotron 3.5 Super VL benchmarks
| Agentic benchmark | Nemotron 3.5 Super VL | Nemotron 3 Ultra |
|---|---|---|
| Hermes Bench (Nvidia-reported) | 63.44% | 56.33% |
| LangChain behavioural evals (Nvidia-reported) | 87.05% | 84.3% |
| BrowseComp | 78 | 67.85 |
| Terminal-Bench 2.1 | 60.67 | 54.80 |
| SWE-bench Pro | 51.78 | 49.66 |
| OSWorld-Verified | 51.14 | N/A |
Full tables, including vision and coding, are in my Nemotron 3.5 Super VL benchmark breakdown.
How to use Nemotron 3.5 Super VL
| Where | How | Cost |
|---|---|---|
| OpenCode | Pick Nemotron 3.5 Super VL in the model drop-down | Free (at time of writing) |
| OpenRouter | Model id nvidia/nemotron-3.5-super-vl or :free | Free variant, or $0.15/M input and $0.50/M output tokens |
| Hermes Agent | Choose it via OpenRouter (or your local server) in the model picker | Free or pay-as-you-go |
| Ollama | ollama run nemotron-3.5-super (80–90GB) | Free, needs big hardware |
| LM Studio | Download the GGUF (Q4_K_M, about 70GB) | Free, needs 70GB+ RAM |
| Hugging Face | BF16 and NVFP4 checkpoints for vLLM or SGLang | Free weights, server GPUs |
Is Nemotron 3.5 Super VL any good?
Let me be clear: this is not a frontier model. It’s not going to replace Claude Opus 5.5 for your hardest work. What it is: a free, open, fast model that can read screenshots, PDFs and charts and then act on them — which is exactly the job a lot of agent work starts with.
In my test I asked it in OpenCode to build an entertaining to-do list app, and it built a working one quickly. For jobs that start with a picture or a file — reading client reports, receipts, screenshots — it’s a great model to slot into your agent.
What Nvidia released with it
This is one of the most open releases from a big lab. Alongside the model weights, Nvidia published training recipes, reinforcement-learning environments and open datasets covering agentic tool use, long context and vision-language understanding. The model was trained on about 26 trillion tokens and post-trained with roughly 3.8 million samples, according to Nvidia. That means companies can fine-tune it on their own way of working, rather than just using it as-is.
Nvidia also launched two smaller models at the same time: Bolt 74B-A9B, which fits in 24GB of memory, and Bolt E4B, a 4GB model for fast conversational agents.
Where it fits in your agent stack
Think of it as your agent’s eyes. Route any job that starts with a picture or a file — a client report, a receipt, a screenshot, a chart — to Nemotron 3.5 Super VL, keep a fast text model for quick research, and save a frontier model for the hardest reasoning. Because it’s free in OpenCode and on OpenRouter, adding it costs nothing, and it fills a gap most free agent setups have.
My verdict on Nemotron 3.5 Super VL
Nemotron 3.5 Super VL is one of the most useful free models for agents right now: open, fast and able to see. Use it free in OpenCode or Hermes, and keep a frontier model for the hardest tasks.
FAQ: nemotron 3.5 super vl
What is Nemotron 3.5 Super VL?
Nvidia’s open 120B-parameter (12B active) vision-language model for agents, released 15 October 2026, which reads text, images, documents and short videos.
Is Nemotron 3.5 Super VL free?
Yes — it’s free in OpenCode at the time of writing, has a free variant on OpenRouter, and its weights are open on Hugging Face.
Is it better than Nemotron 3 Ultra?
On Nvidia’s agent tests it scored higher — 63.44% vs 56.33% on Hermes Bench — despite being smaller.
Can I run it on my computer?
Only with a lot of memory: the Ollama build is 80–90GB and LM Studio’s smallest is about 70GB.
Is it a frontier model?
No — it’s a fast, free open model, not a replacement for Claude Opus 5.5.
Two ways I can help from here. Join the AI Profit Boardroom — it’s where your AI agent model setup gets built with 3,400+ members doing the same.
Or grab a free strategy session and bring your questions.
Join the AI Profit Boardroom →Claim Your Free SEO Strategy Session →
About Julian Goldie
I’m Julian Goldie — founder of Goldie Agency, best-selling author, and an AI educator with 428K+ YouTube subscribers. I’ve spent 10+ years in SEO and online business and I test every AI agent on my own work before I recommend it. Join the AI Profit Boardroom for the daily builds, or book a free strategy session to talk through yours.
Related reading
Last updated October 2026. This page is a living guide to nemotron 3.5 super vl — the facts here move fast and it is updated as they do.