Nvidia has introduced new open-weight models aimed at two settings: enterprise agents running on company infrastructure and AI assistants operating on local computers. The lineup includes a large model for business tasks and two lighter Bolt models designed for PCs and workstations.
A larger model for enterprise agents
Nemotron 3.5 Super VL is a 120-billion-parameter model intended for enterprise agents. Nvidia says it has fewer parameters than Nemotron 3 Ultra while offering similar accuracy on instruction-following and tool-use tasks.
The company describes Super VL as suited to specialized agent work. Its examples include document intelligence, financial accounting, scientific experimentation, medical insurance claims, and security attack resilience. Nvidia also claims that, when post-trained for specific agentic tasks, Super VL completes tasks faster than competing models from Anthropic and OpenAI (Opus 5, GPT 6.1 Sol, Luna, Astra).
Those performance comparisons are Nvidia’s claims. The company also points to organizations it says are using Super VL: Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce.
Local models for PCs and workstations
Nvidia’s other new releases are two lightweight Bolt models. They are intended to run locally on AI PCs using the Nvidia RTX Spark chip and on workstations running DGX Spark. Nvidia says these machines are being released through partners such as Microsoft and Dell for consumers, programmers, and enterprise knowledge workers.
The larger of the two, Nemotron 3.5 Bolt 74B-A9B 24GB, is a 74-billion-parameter model for local agentic tasks. Nvidia says it is designed to balance speed with reasoning across multistep reasoning, tool calling, and web browsing. The smallest model, Bolt E4B, is aimed at conversational agents.
Super VL and Bolt 74B-A9B are available now, while Bolt E4B is expected later this month, according to Nvidia.
Why Nvidia is emphasizing open models
Organizations can post-train Super VL on their own data and deploy it on their own infrastructure, according to Nvidia’s vice president of Open Models and Software, Kari Briski. That approach gives companies a way to build and operate agents using a model they can adapt and run themselves.
The push comes as enterprise companies show interest in running open models in their own data centers. Nvidia has previously released other open-weight models in its Nemotron family, spanning multimodal artificial intelligence, reasoning, coding, math, and enterprise automation. At the same time, AI labs are looking to chip and system providers beyond Nvidia, including AWS’s Trainium and Cerebras, and developing their own chips.
Local assistants and privacy
The Bolt models also connect to Nvidia’s effort to advance consumer AI assistants. Assistants may handle information about family, health, or finances, and some consumers may be wary of granting them access to that personal information.
Running models on a local PC offers a different deployment choice. Nvidia says users can delegate everyday projects without sending those workloads to the cloud. The company is therefore promoting both the models and the PCs powered by its chips as parts of a local-agent setup.
Together, the releases address different priorities: Super VL targets enterprise capability and deployment on company infrastructure, while Bolt models are designed for local computers. Nvidia’s pitch is that organizations and PC users can choose models suited to their needs and where they want the work to run.