Nvidia launches new open models to run agents in the enterprise and on PCs
techcrunch ·
Nvidia has launched new open-weight models designed to support always-on agents for both enterprise environments and edge computing devices. The Nemotron 3.5 Super VL is a large 120-billion-parameter model intended for corporate use, which Nvidia claims delivers accuracy comparable to its Ultra variant while offering faster task completion than competitors like Anthropic and OpenAI. On the consumer side, the company introduced lightweight Bolt models, including the 74-billion-parameter Bolt 74B-A9B and the smaller E4B, which are optimized to run locally on AI PCs equipped with Nvidia RTX Spark chips. These local models allow users to process data without sending workloads to the cloud, addressing privacy concerns associated with cloud-based AI assistants. Major organizations such as Salesforce, CrowdStrike, and EY are reportedly already utilizing the Super VL model for specialized tasks like financial accounting and security resilience. Nvidia emphasizes that these open models enable enterprises to post-train systems on their own data and maintain full ownership of their infrastructure. This strategic move aims to capture market share in both the growing enterprise automation sector and the emerging consumer AI assistant space. The Super VL and Bolt 74B-A9B models are currently available, with the Bolt E4B scheduled for release later this month.
techcrunch