Nvidia is widening its bet that the next wave of artificial intelligence runs as always-on agents, not chat windows. The company on Monday launched a family of open-weight models built for enterprise agents and edge computing, headlined by Nemotron 3.5 Super VL, a 120-billion-parameter model aimed at document intelligence, financial accounting, medical insurance claims and security work.
Nvidia says Super VL delivers instruction-following and tool-use accuracy comparable to its larger Nemotron 3 Ultra, and completes post-trained, task-specific work faster than rival models from Anthropic and OpenAI. Early users named by the company include Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce.
Alongside the enterprise model, Nvidia announced two lightweight Bolt models designed to run locally on a new class of AI PCs built around its RTX Spark chip and DGX Spark workstations, through partners including Microsoft and Dell; a Bolt E4B variant is scheduled for October 23.
“Organisations can run agents on Nvidia infrastructure without having to choose between capability and deployability,” said Kari Briski, a company vice-president. The launches underline Nvidia’s strategy of selling not just the chips agents run on, but the models that run on them.
