Nvidia launches new open models for enterprise data centers and local AI PCs

Nvidia has launched a new set of open-weight models aimed at powering autonomous agents for both large businesses and personal computers. The headline release is the Nemotron 3.5 Super VL, which is a massive 120-billion-parameter model built specifically for enterprise tasks.

Enterprise Tools

Nvidia states that the Super VL model can deliver instruction-following and tool-use accuracy similar to its larger Nemotron 3 Ultra model while using fewer parameters. The company claims the model beats competing systems from Anthropic and OpenAI in speed on specific tasks after post-training. These specific tasks cover document intelligence, financial accounting, scientific experimentation, medical insurance claims, and security attack resilience. Major companies already using Super VL include Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce.

Local AI PCs

On the consumer side, Nvidia also introduced two lightweight Bolt models intended to run locally on new AI PCs and workstations. These machines run on the Nvidia RTX Spark and DGX Spark chips and are rolling out now through partners like Microsoft and Dell. The larger of the two is the Nemotron 3.5 Bolt 74B-A9B 24GB, a 74-billion-parameter model built for multistep reasoning, tool calling, and web browsing. The smaller Bolt E4B model is designed for conversational agents. Super VL and the larger Bolt are available now, while the Bolt E4B launches later this month.

Context

Open-weight models allow organizations to download and modify software code to fit their specific operational needs rather than relying strictly on cloud-hosted services. Running these systems locally on personal computers addresses growing consumer concerns about privacy by keeping personal data off external servers. Nvidia's hardware rollout directly targets the growing market for specialized AI processing chips across both massive corporate data centers and individual desktop computers.