Key points
- Nvidia released Nemotron 3.5 Super VL, a 120-billion-parameter open-weight model built for enterprise agents, available now.
- Two smaller Bolt models target local use on RTX Spark AI PCs and DGX Spark workstations; the 74B-A9B is available now, the E4B later this month.
- Nvidia says Super VL, once post-trained, finishes agentic tasks faster than rival models from Anthropic and OpenAI. Named users include Salesforce, EY and Perplexity.
Why it matters for marketers: Open models that run on a company's own servers or on local PCs give marketing teams a route to agents that keep customer and brand data in-house.
Nvidia has launched new open-weight models for always-on agents. Nemotron 3.5 Super VL is a 120-billion-parameter model aimed at enterprise use. Nvidia says it matches the larger Nemotron 3 Ultra on instruction-following and tool use despite having fewer parameters.
The company also introduced two lightweight Bolt models meant to run locally on new AI PCs with the RTX Spark chip and on DGX Spark workstations. Bolt 74B-A9B handles multistep reasoning, tool calling and web browsing, while the smaller E4B targets conversational agents. Super VL and Bolt 74B-A9B are available now, and E4B is due later this month.
Nvidia claims Super VL, after post-training on specific tasks, completes work faster than competing models from Anthropic and OpenAI, including document intelligence and financial accounting. Those are vendor claims, though Nvidia cited Skydance, EY, Ramp, Crowdstrike, Perplexity and Salesforce as users. Nvidia's Kari Briski said organizations can post-train Super VL on their own data and "own what they build."
For marketers and content teams, the appeal is control. Agents that run on a company's own infrastructure, or on a laptop without sending work to the cloud, can ease privacy and data-handling concerns around customer information, drafts and campaign plans. Agencies handling client data may find the local option especially relevant.
Watch for independent benchmarks, since the speed claims come from Nvidia. Also check how the models perform on your own content workflows after post-training, and what hardware and tuning effort a self-hosted setup would require.
Sources
This brief is our own summary. Read the full reporting at the source.
