Nvidia Nemotron 3.5 models are expanding the company’s open-weight AI lineup with three releases designed for enterprise automation and local computing. The new offerings include a 120-billion-parameter model for demanding business workflows and two smaller models built for PCs and edge devices.
In its October 15 announcement, Nvidia introduced Nemotron 3.5 Super VL alongside Bolt 74B-A9B and Bolt E4B. The company says the models give developers greater control over AI agents without relying entirely on proprietary cloud services.
Nemotron 3.5 Super VL Targets Enterprise AI Agents
Super VL is designed for organizations running specialized AI agents continuously. With 120 billion parameters, it is substantially smaller than Nemotron 3 Ultra, Nvidia’s 550-billion-parameter reasoning model. Nvidia claims Super VL delivers comparable accuracy on instruction-following, tool use and complex business workflows.
The company also claims Super VL completes specialized tasks up to five times faster than competing open models in its class. These performance figures reflect Nvidia’s reported benchmarks and may vary across workloads.
Companies adopting or customizing Super VL include Salesforce, CrowdStrike, Perplexity, Ramp, EY and Skydance. Their applications span financial document analysis, cybersecurity, scientific research and enterprise automation. Salesforce has already worked with Nvidia on Koa, an enterprise reasoning model built using Nemotron technology.
Unlike closed proprietary systems, Super VL lets organizations customize models with their own data and run them on controlled infrastructure. Nvidia provides open weights, training resources and deployment tools through its Nemotron developer platform.
Bolt Models Bring AI Agents to Local PCs
Nvidia also introduced two smaller models intended for local computing. Nemotron 3.5 Bolt 74B-A9B 24GB uses memory-efficient quantization to run within 24GB, supporting multistep reasoning, coding and web research.
The model supports Nvidia RTX Spark laptops, GeForce RTX PCs and DGX Spark workstations. Its release coincides with the arrival of RTX Spark-powered laptops from manufacturers including Microsoft and Dell.
The smaller Bolt E4B 4GB targets conversational assistants, interactive characters and embedded AI systems. It supports hardware including GeForce RTX, Jetson and RTX Spark devices.
Local execution offers an alternative for users concerned about sending personal documents, financial information or other sensitive data to cloud services. However, privacy also depends on application permissions and how the software handles information.
The launch strengthens Nvidia’s effort to make its hardware a platform for customizable AI agents across data centres and personal computers. Super VL and Bolt 74B-A9B are available now, while Bolt E4B is scheduled for October 23.
