AI models · NVIDIA

Nemotron 3.5 Super VL

Nemotron 3.5 Super VL came out on 15 October 2026 and is built for AI agents that work with tools, documents and screenshots. It has 120 billion parameters, but only 12 billion are active for each word, so it answers quickly. The weights are free to download and the license allows commercial use. It suits companies that want their own agent on a server or a cheap API for large volumes.

Strengths

  • Open weights under the OpenMDW 1.1 license. It allows commercial use, changes and redistribution with no geographic restrictions, as long as you pass on the license and original notices.
  • Speed: Artificial Analysis measured 447 tokens per second, and NVIDIA says it is up to five times faster than the next best open model in its class.
  • It understands text, images, documents, screenshots and videos up to 2 minutes long, with a context of up to 1 million tokens.
  • NVIDIA says it is comparable to its largest reasoning model on instruction following, tool use and visual understanding, with about a fifth of the parameters.
  • Low price from API providers: on OpenRouter typically $0.15 per million input tokens and $0.50 per million output tokens.
  • Reasoning can be switched on, off or set to medium, depending on whether you want speed or accuracy.

Weaknesses

  • All quality test results come from NVIDIA. It reports 71.1% on the SWE-bench Verified coding test, below Beam 1.0 (81.6% according to Reflection AI).
  • The full version needs a server with four H100 or two GB200 GPUs. The leaner NVFP4 version runs on a single H200, an RTX PRO 6000 or a DGX Spark, but an ordinary PC or laptop will not do.
  • Slovak and Czech are not among the languages NVIDIA fine-tuned it on. Test it before using it for texts in those languages.
  • NVIDIA publishes no per-token price. Under its terms, the trial access on build.nvidia.com is for testing only, not production, and you may not send personal or confidential data to it.
  • Many API providers offer only a 262,144 token context, not the full 1 million.

Best for

  • business AI agents that use tools
  • reading documents, tables and screenshots
  • cheap high-volume tasks through the API
  • a multimodal assistant on your own server

Not ideal for

  • laptops and ordinary PCs
  • the hardest coding tasks
  • customer-facing texts in Slovak or Czech without testing first

How to use it in Europe

The weights are free to download from Hugging Face anywhere in Europe, and the OpenMDW 1.1 license has no geographic restrictions. On your own hardware (the NVFP4 version needs just one H200 or a DGX Spark), your data stays in house, which is the safest route for GDPR. Through an API, providers such as Together, DeepInfra and CoreWeave offer it via OpenRouter. Check with each provider where it processes data. NVIDIA, a US company, only offers trial access on build.nvidia.com for testing, not for production or personal data.

Price compared with similar models

1 3 10 Nemotron 3.5 Super VL0.5 Llama 4 Maverick0.65 Qwen3.7-Plus1.6 Gemini 3.8 Flash3.75 Muse Spark 1.34.25 Muse Spark Open4.25 Mistral Medium 3.57.5 Claude Sonnet 5.510 GPT-6.1 Sol10
Output price in USD per 1M tokens (log scale). Lighter bars are open-weight models.

Benchmark results

BenchmarkScoreSource
GPQA Diamond89.52%huggingface.co ↗ (vendor-reported), October 2026
SWE-bench Verified71.10%huggingface.co ↗ (vendor-reported), October 2026
BrowseComp78%huggingface.co ↗ (vendor-reported), October 2026
OSWorld-Verified51.14%huggingface.co ↗ (vendor-reported), October 2026
Artificial Analysis output speed447 tokens/sartificialanalysis.ai ↗, October 2026

Benchmarks cover only a narrow slice of ability and vendors often report them at their best settings. Treat them as a rough guide.

More models from this vendor

  • Nemotron 3.5 Bolt 74B-A9B: NVIDIA's open model for AI agents on your own powerful PC, small enough to fit in 24 GB of graphics card memory.

Similar models from other vendors

  • Claude Sonnet 5.5 (Anthropic): Claude's balanced model for everyday work and coding. It is faster than Opus 5.5 and costs half as much.
  • Command A+ (Cohere): An open business model from Canada's Cohere that officially supports Slovak and Czech and runs on your own server.
  • Gemini 3.8 Flash (Google): Google's fast, lower-cost model for everyday work, coding and document analysis. You can try it in the API for free.
  • Muse Spark 1.3 (Meta): Meta's balanced model with a 1-million-token window and a fair price. The Muse Spark line also powers the Meta AI assistant.
  • Muse Spark Open (Meta): Meta's large open model that a company can run on its own servers, so data never leaves its infrastructure.
  • Llama 4 Maverick (Meta): The larger model in the newest Llama generation, from April 2025. Cheap through providers, but its license excludes people and companies in the EU.

Sources

Prices and models change often. The data comes from official vendor pricing pages and documentation, and every page shows when it was last checked. Before a larger purchase, confirm the price with the vendor.

← All AI models

Not sure which model fits your business?

I can help you pick a model and plan based on what you actually need, set it up safely for company data and train your team to use it well. Let’s start with an intro consultation.

See pricing [email protected]