Nvidia launched three open-weight models on Thursday designed to run AI agents on enterprise infrastructure and consumer PCs without relying on cloud services.
The Nemotron 3.5 Super VL, a 120-billion-parameter model, targets enterprise deployments. Nvidia claims it matches the accuracy of its larger Nemotron 3 Ultra predecessor on instruction-following and tool-use tasks. The company says Super VL completes tasks faster than competitors' models including Anthropic's Opus 5 and OpenAI's GPT 6.1 Sol on document intelligence, financial accounting, scientific experimentation, medical insurance claims, and security tasks. Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce use Super VL, according to Nvidia.
Two smaller Bolt models target local processing on consumer and business PCs. The Nemotron 3.5 Bolt 74B-A9B, a 74-billion-parameter model, balances speed with reasoning for multistep tasks, tool calling, and web browsing. The Bolt E4B, designed for conversational agents, arrives later this month. Both run on Nvidia's RTX Spark chip in new AI PCs from Microsoft and Dell and on DGX Spark workstations.
Kari Briski, Nvidia's vice president of Open Models and Software, framed the releases as giving organizations flexibility. "Any organization that wants to run agents on Nvidia infrastructure now has an open, optimized model for their tier without having to choose between capability and deployability," Briski wrote. Organizations can post-train Super VL on proprietary data and deploy it internally, avoiding vendor lock-in.
The releases reflect pressure on Nvidia as frontier AI labs like Anthropic and OpenAI diversify hardware suppliers. AWS's Trainium, Cerebras, and in-house chips now power competing models. Nvidia counters by pushing enterprises toward running open models on their own infrastructure rather than accessing proprietary cloud services.
The local PC models address privacy concerns. Consumers hesitant to share family, health, and financial data with cloud-based AI assistants can "delegate everyday projects without sending those workloads to the cloud," Briski wrote. Super VL and Bolt 74B-A9B are available now.