Nvidia launches open Nemotron 3.5 models for AI agents

Nvidia has expanded its open-source AI lineup with the release of the Nemotron 3.5 family, featuring open-weight models engineered to run autonomous agents on enterprise infrastructure and personal computers. The flagship Nemotron 3.5 Super VL is a 120-billion-parameter model designed for complex corporate workloads. According to the company, Super VL delivers superior execution speed over competing closed systems like Opus 5, GPT 6.1 Sol, Luna, and Astra when tuned for tasks like medical insurance claims processing, document intelligence, financial accounting, scientific experimentation, and security attack resilience.

Alongside its flagship enterprise model, Nvidia introduced lightweight architectures designed to operate locally on consumer PCs and workstations powered by Nvidia RTX Spark chips and DGX Spark hardware. The Nemotron 3.5 Bolt 74B-A9B 24GB model uses 74 billion parameters to balance speed with multistep reasoning, web browsing, and tool interaction. A smaller conversational model, the Bolt E4B, will be released later this month. Hardware partners Microsoft and Dell are already shipping compatible PCs to support these decentralized setups.

The deployment comes as frontier labs increasingly explore alternative hardware options, including Cerebras and AWS's Trainium chips. By offering customizable weights, Nvidia is positioning its hardware around organization-owned deployments. Kari Briski, Nvidia's vice president of Open Models and Software, noted that these open frameworks allow companies to train models on proprietary data and retain complete ownership over their infrastructure. Early enterprise adopters of Super VL include Salesforce, Crowdstrike, Perplexity, EY, Ramp, and Skydance.

For practitioners and enterprise architects, these additions reduce dependence on closed cloud APIs. Developers can now fine-tune powerful agentic systems directly on internal hardware without exposing sensitive corporate records, financial data, or private user information to third-party providers.