Nvidia has released new open-weight AI models for software agents that run in company data centers, workstations and personal computers. The release gives businesses and developers more options to customize agents and keep workloads on local hardware instead of sending them to the cloud.
120 Billion Parameters
The main model, Nemotron 3.5 Super VL, has 120 billion parameters. It is designed for agents that follow instructions, use tools, read visual documents and complete multistep business tasks. Nvidia says it delivers accuracy comparable to its larger Nemotron 3 Ultra model and is up to five times faster than the next-best open model in its class.
Super VL includes open weights, datasets and training recipes. Organizations can post-train it on their own data and deploy it on infrastructure they control. Nvidia lists finance, insurance claims, cybersecurity, software development, engineering and scientific research among its intended uses.
Bolt Brings Agents Local
Nvidia also released two smaller Bolt models for local systems. Bolt 74B-A9B 24GB has 74 billion parameters and is designed for multistep reasoning, tool calls and web browsing on Nvidia RTX systems and DGX Spark computers. The smaller Bolt E4B is aimed at conversational and embedded agents, including local voice assistants.
Nvidia says the models can handle everyday projects without sending the related workloads to the cloud. Bolt 74B-A9B is available now. Bolt E4B is scheduled for October 23, 2026, according to Nvidia's announcement.
Enterprise Adoption Expands
TechCrunch reported that Nvidia identified Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce as companies using Super VL. Nvidia's announcement also names Atlassian, Dassault Systèmes and Synopsys as organizations integrating the model into search, engineering and semiconductor design workflows.
Nvidia says Super VL runs across its hardware stack, including DGX Spark, DGX Station, RTX systems, Jetson devices, data centers and cloud environments. The company also says the model is available through Hugging Face, OpenRouter, ModelScope and NVIDIA NIM.
References
Bort, J. (2026, October 19). Nvidia launches new open models to run agents in the enterprise and on PCs. TechCrunch. https://techcrunch.com/2026/10/19/nvidia-launches-new-open-models-to-run-agents-in-the-enterprise-and-on-pcs/
Briski, K. (2026, October 15). Open from data center to desktop: NVIDIA expands Nemotron family with highest efficiency Super VL and Bolt models. NVIDIA Blog. https://blogs.nvidia.com/blog/nemotron-3.5-super-vl/