Nvidia releases open Nemotron 3.5 Super VL and Bolt models for enterprise agents and PCs

Nvidia on Thursday launched new open-weight models built for always-on agents and edge computing. The largest, Nemotron 3.5 Super VL, has 120 billion parameters and is aimed at agents used in enterprises. Nvidia says that although it uses fewer parameters than the Nemotron 3 Ultra model, it delivers similar accuracy on instruction-following and tool-use tasks.

Nvidia also released two lightweight Bolt models meant to run locally on new AI PCs built around the Nvidia RTX Spark chip, and on workstations running the DGX Spark. Those PCs, aimed at consumers, programmers and enterprise knowledge workers, are reaching the market through partners such as Microsoft and Dell.

Of the Bolt models, Nemotron 3.5 Bolt 74B-A9B 24GB has 74 billion parameters and is designed for local agentic work, balancing speed with reasoning for multistep tasks, tool calling and web browsing. The smaller Bolt E4B targets conversational agents. Super VL and Bolt 74B-A9B are available now; Bolt E4B follows later this month.

Nvidia claims Super VL completes tasks faster than competing models from Anthropic and OpenAI when post-trained on specific agentic tasks, including document intelligence, financial accounting, scientific experimentation, medical insurance claims and security attack resilience. It lists Skydance, EY, Ramp, Crowdstrike, Perplexity and Salesforce as users.

The release continues Nvidia's push into open-weight models that compete with the proprietary labs, which are both its customers and its investments. Those labs are now sourcing compute from other chip and system makers and developing their own silicon, which weakens Nvidia's hold on them. Courting enterprises that want to run models in their own data centers is one way to offset that.

Nvidia's pitch is control: Kari Briski, its vice president of Open Models and Software, says organizations can post-train Super VL on their own data, deploy it on their own infrastructure and own what they build. The speed comparisons with Anthropic and OpenAI models come from Nvidia and apply to post-trained setups, so they should be read as vendor claims.

The Bolt models serve a second goal: creating a market for PCs that run agents locally. Nvidia's argument is that users can delegate everyday projects without sending workloads to the cloud. Whether privacy-wary consumers will give assistants access to family, health and financial information is an open question.

Watch for the Bolt E4B release later this month and for independent benchmarks that test Nvidia's speed and accuracy claims against the Anthropic and OpenAI models it cites.

Also worth tracking is how quickly RTX Spark PCs from Microsoft, Dell and others sell, since consumer demand for local agents will determine whether the Bolt models find an audience.