A two-tier open-model launch
Nvidia used a Thursday release to draw a clean line between its data-center and desktop strategies. The flagship Nemotron 3.5 Super VL is a 120-billion-parameter open-weight model built for always-on enterprise agents. It has fewer parameters than the older Nemotron 3 Ultra, but Nvidia claims comparable accuracy on instruction-following and tool use, thanks in part to a mixture-of-experts design that activates roughly 12B parameters per token.
At the other end of the spectrum are two lightweight Bolt models aimed at local hardware. Nemotron 3.5 Bolt 74B-A9B 24GB carries 74 billion parameters and is tuned for multistep reasoning, tool calling, and web browsing. Bolt E4B, arriving later this month, targets conversational agents. Super VL and the 74B Bolt are available now.
Beating closed frontier labs on enterprise jobs
TechCrunch reports that Nvidia claims Super VL, after post-training on specific agentic tasks, completes work faster than Anthropic's Opus 5 and OpenAI's GPT 6.1 Sol, Luna, and Astra. The workloads Nvidia calls out are a direct hint at where enterprises want to keep data in-house:
- Document intelligence
- Financial accounting
- Scientific experimentation
- Medical insurance claims
- Security attack resilience
Reference customers named by Nvidia include Skydance, EY, Ramp, CrowdStrike, Perplexity, and Salesforce. Kari Briski, Nvidia's VP of Open Models and Software, framed the launch as removing a trade-off: organizations shouldn't have to "choose between capability and deployability." The strategic subtext is that frontier AI labs are diversifying onto AWS Trainium, Cerebras, and homegrown silicon, so Nvidia wants open weights that keep its GPUs at the center of enterprise AI stacks.
A local privacy play for AI PCs
The Bolt family is designed to run on new AI PCs using RTX Spark chips and DGX Spark workstations, which are shipping now through partners including Microsoft and Dell. The 74B-A9B balances speed with reasoning for agentic workloads, while E4B is leaner for everyday chat. Nvidia's pitch is privacy: delegate routine projects to a local agent without sending family, health, or financial data to the cloud. By making local agent hardware compelling, Nvidia hopes to create a new market for its chips among consumers and knowledge workers who are weary of cloud-based assistants.
Nvidia on Thursday launched new open-weight models designed for always-on agents and edge computing.
The Nemotron 3.5 Super VL is a large, 120-billion-parameter model geared for agents used by the enterprise. While it uses fewer parameters than the Nemotron 3 Ultra model, Nvidia says it can deliver similar accuracy in instruction-following and tool-use tasks.
At the other end of the spectrum, the company also launched two lightweight Bolt models. These are intended to run locally on the new AI PCs (running the Nvidia RTX Spark chip) and workstations running the DGX Spark. Such PCs, aimed at consumers, programmers and enterprise knowledge workers, are being released right now into the market through partners like Microsoft and Dell.