Nvidia releases open AI models for business agents and PCs

In short: Nvidia has released new “open-weight” AI models that companies can run in their own data centers and that some PCs can run locally.

What happened

Nvidia announced a new set of AI models under its Nemotron line. The biggest one is called Nemotron 3.5 Super VL. Nvidia says it is designed for “agents,” which are AI helpers that can do tasks like reading documents, using software tools, and following step by step instructions (like a digital assistant that can actually carry out work, not just answer questions).

Nvidia also launched two smaller models called Bolt. These are meant to run on certain new “AI PCs,” including systems using the Nvidia RTX Spark chip, plus DGX Spark workstations. “Run locally” means the AI can work on your own computer instead of sending the request to a company server on the internet, which is often called “the cloud” (think of the cloud as renting a powerful computer somewhere else).

Nvidia said Super VL can be trained further on a company’s own data and then deployed on that company’s own computers. Nvidia also claimed Super VL finishes some agent style tasks faster than competing models from companies like OpenAI and Anthropic, and it listed organizations it says are using Super VL, including Skydance, EY, Ramp, CrowdStrike, Perplexity, and Salesforce.

Super VL and one Bolt model are available now. Nvidia said the smallest Bolt model will arrive later this month.

Why it matters

Many companies and some everyday users want AI that keeps more data private and stays under their control. Nvidia is betting that more people will choose AI that runs on their own hardware, either in a company data center or on a personal computer.

Source: TechCrunch AI