Nvidia has introduced new open-weight models for enterprise agents and edge computing. The flagship Nemotron 3.5 Super VL, a 120-billion-parameter model, is designed for instruction-following and tool-use tasks. Nvidia says it delivers accuracy similar to its larger Nemotron 3 Ultra while using fewer parameters, giving organizations a more efficient option for running agents on their own infrastructure.
For local deployment, Nvidia launched two lightweight Bolt models. The 74-billion-parameter Bolt 74B-A9B 24GB is built for multistep reasoning, tool calling, and web browsing on AI PCs and workstations powered by Nvidia's RTX Spark and DGX Spark chips. A smaller Bolt E4B, aimed at conversational agents, is expected later this month. These systems are reaching the market through partners such as Microsoft and Dell.
Nvidia claims that Super VL completes agentic tasks faster than competing models from Anthropic and OpenAI when post-trained on specific workloads like document intelligence, financial accounting, and security attack resilience. The company lists Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce as users. Unlike closed frontier models, organizations can post-train Super VL on their own data and deploy it in their own data centers, according to Nvidia's vice president of open models and software, Kari Briski.
The release also pushes the trend of consumer AI assistants running locally, addressing privacy concerns about sending personal data to the cloud. Nvidia faces growing competition from custom chips such as AWS Trainium and Cerebras, as well as in-house silicon from AI labs, making open models and PC hardware a strategic bet for the company.