Nvidia unveiled new open-weight AI models on October 19, 2026, focused on running always-on agents in enterprises and on personal computers. The flagship Nemotron 3.5 Super VL, a 120-billion-parameter model, aims to serve large-scale business needs by delivering comparable accuracy to larger models like Nemotron 3 Ultra while using fewer parameters. Additionally, Nvidia introduced two smaller Bolt models designed to operate locally on AI-powered PCs using the Nvidia RTX Spark chip and DGX Spark workstations, which are being marketed by partners including Microsoft and Dell.

The Super VL model targets enterprise applications such as document processing, financial accounting, scientific research, medical claims, and cybersecurity, and Nvidia claims it outperforms competitors like Anthropic and OpenAI in these post-trained agent tasks. Significant companies including Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce are reportedly utilizing the new Super VL model. This launch expands Nvidia’s existing lineup of Nemotron models, which support a variety of AI tasks from reasoning and coding to automation.

Nvidia is positioning these models to meet the growing demand from enterprises that prefer running open models in-house rather than relying on proprietary cloud services. This strategy allows organizations to customize and fine-tune the models on their data, providing them control over AI workflows and infrastructure. Nvidia’s vice president of Open Models and Software, Kari Briski, emphasized that Super VL offers enterprises flexibility without sacrificing performance or deployment ease.

Beyond the enterprise focus, Nvidia is also pushing consumer adoption of AI assistants with the Bolt models. The Bolt 74B-A9B, with 74 billion parameters, balances speed and reasoning for tasks like tool integration and web browsing, while the smaller Bolt E4B is optimized for conversational agents and will be released later this month. By enabling AI agents to run locally on PCs, Nvidia aims to address privacy concerns by avoiding constant cloud communication, allowing users to manage daily projects while keeping their data private.