Nvidia releases open 120B Nemotron 3.5 Super VL to read documents, charts and video
Representative image: Magnascan / Pixabay
Nvidia released Nemotron 3.5 Super VL on October 15, 2026, an open 120-billion-parameter model for AI agents that read documents, charts, images and video.
What Happened
Nvidia has released Nemotron 3.5 Super VL, an open multimodal artificial intelligence model with 120 billion parameters that is built for AI agents that read documents, charts, images and short videos, the company said on Thursday, October 15, 2026. The model uses a mixture-of-experts design that activates about 12 billion parameters for each piece of text it processes, supports a context window of up to one million tokens, and is released with its weights, training datasets and recipes under a permissive open licence.
Key Facts
- Model: Nemotron 3.5 Super VL, released by Nvidia on October 15, 2026.
- Size: 120 billion parameters, with about 12 billion active per token (mixture of experts).
- Inputs: text, images, documents and MP4 video clips of up to two minutes; it does not process audio.
- Context window: up to one million tokens.
- Licence: OpenMDW-1.1, a permissive licence; Nvidia is also releasing datasets and training recipes.
- Focus: AI agents that extract information from tables, charts and page layouts.
- Nvidia's example: after post-training on a medical insurance claims task, accuracy rose from about 28% to 72%, according to the company.
Why It Matters
Much of the world's business information is locked in documents: invoices, insurance claims, contracts, financial reports and forms full of tables and charts. Models that can read such material accurately, and then act on it, are what companies need to automate back-office work. Nvidia is positioning Nemotron 3.5 Super VL as a base that companies can download and fine-tune for those specialised tasks, rather than sending sensitive documents to an outside service.
The release also shows how Nvidia, best known for the chips that train and run AI models, is building its own family of open models. By releasing training data and recipes as well as weights, it lets developers see how the model was made and adapt it more easily, and it encourages more AI work to run on its hardware.
There are caveats. Nvidia's headline claim of about five times higher throughput than rivals relies on a leaderboard that had not yet scored the model at launch, one technology outlet noted, and different reports quote slightly different benchmark numbers. Independent tests will show how well it performs outside Nvidia's own examples.
| Item | Detail |
|---|---|
| Total parameters | 120 billion |
| Active parameters | About 12 billion |
| Context window | Up to 1 million tokens |
| Inputs | Text, images, documents, short video |
| Licence | OpenMDW-1.1 |
For Indian companies and start-ups building document-processing tools, in areas such as banking, insurance and government services, an open model of this kind offers an option that can be run on their own infrastructure.
Impact
Short-term: Developers can download the model and test it on their own document and chart tasks. Expect independent benchmarks in the coming weeks.
Long-term: If open multimodal models prove reliable on business documents, more companies may automate claims processing, data entry and report analysis in-house, changing the nature of back-office jobs.
Who is affected: Enterprises with large document workloads; AI developers and start-ups; cloud and GPU providers; and workers in data-entry and processing roles.
Key Takeaway
Nvidia's Nemotron 3.5 Super VL, released on October 15, 2026, is a 120-billion-parameter open model built to read documents, charts and video for AI agents, with weights, data and recipes freely available, though its performance claims still need independent testing.
Questions and Answers
What is Nvidia Nemotron 3.5 Super VL?
It is an open multimodal AI model from Nvidia, released on October 15, 2026, with 120 billion parameters, about 12 billion of them active per token. It is designed for AI agents that work with documents, charts, images and short videos.
Is Nemotron 3.5 Super VL free to use?
It is released under the permissive OpenMDW-1.1 licence, and Nvidia is publishing the model weights, datasets and training recipes.
What can the model process?
It reads text, images, documents and MP4 video clips of up to two minutes, with a context window of up to one million tokens. It does not process audio.
How good is it?
Nvidia reports large gains after task-specific training, such as accuracy on a medical claims task rising from about 28% to 72%. Some of its comparative claims had not been independently verified at launch.
Disclaimer: Prepared by the Peepals newsroom from publicly available sources with AI assistance. Information is accurate to the best of our knowledge at the time of publication and may change. Images may be representative. Not professional advice. Report an error via our contact page.
Sourced and fact-checked by the Peepals Global Editorial Team
NuSynth Technologies raises Rs 36.5 crore to use AI to redesign chemical making
Anthropic commits 5% of shares, about $48 billion, to new public benefit fund
Reported, fact-checked and published by the Peepals Global Editorial Team.
