Nvidia releases open Nemotron models for enterprise agents and local AI PCs

Nvidia has released three open-weight models built for AI agents: Nemotron 3.5 Super VL and two smaller Bolt models, according to TechCrunch.

Super VL has 120 billion parameters and is aimed at agents in large organisations. Nvidia says it matches the larger Nemotron 3 Ultra on following instructions and using tools, with fewer parameters. The Bolt models run locally on AI PCs with Nvidia's RTX Spark chip, sold through partners such as Microsoft and Dell, and on DGX Spark workstations. Bolt 74B-A9B has 74 billion parameters and handles multistep reasoning, tool calling and web browsing. Bolt E4B is meant for conversational agents. Super VL and Bolt 74B-A9B are available now, and E4B follows later in October.

Several AI labs that buy Nvidia chips now also use AWS Trainium or Cerebras, and some are building their own. Open models that companies can train on their own data and run in their own data centres, or on a PC without the cloud, give customers another reason to stay with Nvidia hardware.