Nvidia releases open Nemotron 3.5 models for enterprise agents
Nvidia has introduced three open-weight Nemotron 3.5 models for agentic AI workloads, spanning enterprise data centres and local AI PCs. The 120-billion-parameter Nemotron 3.5 Super VL is aimed at always-on enterprise agents, while Nemotron 3.5 Bolt 74B-A9B 24GB and Bolt E4B target local deployments on RTX Spark-based AI PCs and DGX Spark workstations.
Super VL and Bolt 74B-A9B are available now. Nvidia says Bolt E4B, its smallest new model and a model intended for conversational agents, will arrive later this month. The launches extend Nvidia's open-model portfolio as enterprises weigh how much AI processing and data handling should remain under their own control.
A model tier for enterprise agent tasks
Nemotron 3.5 Super VL has fewer parameters than Nvidia's Nemotron 3 Ultra, but Nvidia says it offers similar accuracy on instruction-following and tool-use tasks. The company positions it for agents used in document intelligence, financial accounting, scientific experimentation, medical insurance claims and security attack resilience.
Nvidia also says Super VL can complete post-trained agentic tasks faster than competing models it names from Anthropic and OpenAI. The company cited Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce as users of Super VL. These are Nvidia's performance and adoption claims, rather than independently reported comparative results.
The release complements a broader discussion of model choices, as open and closed AI model trade-offs examines the trade-offs between open and closed AI models for organizations building AI systems. Nvidia's stated proposition is that customers can post-train Super VL on their own data, deploy it on their own infrastructure and retain ownership of what they build.
Local models for AI PCs and workstations
Bolt 74B-A9B 24GB is a 74-billion-parameter model for local agentic work. Nvidia says it is designed to balance speed and reasoning for multistep reasoning, tool calling and web browsing. Its intended platforms are AI PCs using Nvidia RTX Spark chips and workstations based on DGX Spark.
Those systems are being brought to market through partners including Microsoft and Dell, with consumers, programmers and enterprise knowledge workers among the intended users. Nvidia frames local execution as a way for PC users to delegate everyday projects without sending those workloads to the cloud, a proposition that addresses users wary of sharing personal, health or financial information with cloud-based assistants.
Deployment choice becomes part of the AI decision
Nvidia is releasing the models as proprietary frontier labs, which are both customers and investment targets for the company, explore other chips and AI system suppliers, including AWS Trainium and Cerebras, alongside internally developed chips. At the same time, Nvidia is promoting its own infrastructure for organizations that want to operate open models in their data centres.
For businesses, the practical implication is to match the model tier and deployment location to the workload: assess whether enterprise agents require Super VL's scale, whether local Bolt models meet a task's needs, and whether data, infrastructure and operational requirements support self-hosted deployment.