Nvidia releases free open models for reading charts and PDFs and for business data

What Nvidia announced

As technews.tw reported, Nvidia announced on November 5 a series of free, open AI models and tools. The stated goal is to help developers bring AI into scientific research, engineering analysis and business applications more quickly.

The release is not one model. It spans several kinds of work, from understanding images to reasoning over structured data, and it extends Nvidia's push into open models.

A model that reads screenshots and PDFs

The vision piece is Nemotron 3.5 Super VL. It is built to analyze screenshots, PDFs and charts, the kind of mixed documents that fill most offices. The report says it runs faster than the previous version and performs better, but it gives no benchmark figures.

You can use it for free through OpenCode, OpenRouter, Hugging Face, Ollama and LM Studio. These are common places to try models, either in the cloud or on your own computer, so testing it takes little setup.

Agents that use company data in place

Nvidia also expanded its open-source Agent Toolkit with a feature called Knowledge Foundation. An AI agent is a program that takes steps on a task for you, not just answers questions. This feature lets a company's agents reason directly from internal data without copying that data elsewhere.

Nvidia says companies keep control over what the agents can access and what actions they can take. That matters for any business wary of sending sensitive files out of its own systems.

Kumo models for business data

The third piece is the Kumo model family. It targets tabular data (rows and columns), relational data (linked tables), time series (numbers tracked over time) and forecasting. These match everyday business questions, such as which customers might leave or what demand will look like next quarter.

What the report leaves out

The report says nothing about license terms, model sizes or hardware needs. Those details decide whether you can use a model in a commercial product and whether it runs on your own machine. The report also comes from a secondary outlet, not Nvidia's own page, so the model cards are the place to confirm them.