Nvidia releases open Nemotron models for agents on servers and on PCs

Nvidia launched new open-weight models built for always-on agents. Nemotron 3.5 Super VL is a 120 billion parameter model aimed at enterprise agents. It uses fewer parameters than Nvidia's Nemotron 3 Ultra, but Nvidia says it delivers similar accuracy on instruction following and tool use.

At the other end, two lightweight Bolt models are meant to run locally on new AI PCs and on DGX Spark workstations. The Bolt 74B-A9B model is designed for multistep reasoning, tool calling and web browsing. A smaller Bolt E4B for conversational agents is due later this month. Super VL and Bolt 74B-A9B are available now.

Nvidia claims Super VL finishes tasks faster than rival models from Anthropic and OpenAI when post-trained on specific agentic jobs such as document intelligence, financial accounting and insurance claims. As with any vendor claim, treat it as a starting point. Nvidia lists companies it says use the model, including Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce.

The pitch is that organizations can post-train the model on their own data, run it on their own infrastructure and own what they build.

Our take

Nvidia sells the chips and now also the open models that make companies want to run their own chips. That is a neat loop while the big labs shop for other hardware.