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

Nemotron 3.5 Super VL is a 120B-parameter model, smaller than Nemotron 3 Ultra, but NVIDIA says it approaches the latter's accuracy on instruction-following and tool-calling tasks. Of the two lightweight Bolt models, Nemotron 3.5 Bolt 74B-A9B 24GB has 74B parameters and targets local multi-step reasoning, tool calls and web browsing; the smallest, Bolt E4B, targets conversational agents. Super VL and Bolt 74B-A9B are available now, with Bolt E4B due later this month. NVIDIA says Super VL, after post-training, completes agent tasks faster than Anthropic and OpenAI competitors in document understanding, finance and accounting, scientific experiments, health insurance claims and attack resilience; Skydance, EY, Ramp, Crowdstrike, Perplexity and Salesforce are listed as users. For teams owning agent platforms and data compliance, open weights allow post-training on proprietary data and deployment on owned infrastructure, clearly different from calling closed-model APIs in data ownership and deployment location. The Bolt series targets new AI PCs with RTX Spark chips and DGX Spark workstations now reaching market via partners including Microsoft and Dell. For developers and knowledge workers buying such devices to keep everyday tasks local, workloads need not go to the cloud; NVIDIA is also opening a new shipment scenario for running agents locally on its own chips.