AI News · Models ·
Nvidia releases Nemotron 3.5 open models for enterprise and on-device agents
Nvidia released new open-weight Nemotron 3.5 models built for always-on AI agents in company data centers and on personal computers. The 120 billion parameter Super VL targets enterprise agents, while two Bolt models are designed to run locally on RTX Spark PCs and DGX Spark workstations. Super VL and Bolt 74B-A9B are available now, and Bolt E4B is due later this month.
Key points
- Nemotron 3.5 Super VL has 120 billion parameters and targets enterprise agents.
- Bolt 74B-A9B and Bolt E4B are built to run locally on RTX Spark PCs and DGX Spark.
- Nvidia named EY, Ramp, CrowdStrike, Perplexity, Salesforce and Skydance as Super VL users.
What happened: Nvidia launched a new set of open-weight models aimed at AI agents that run inside enterprises and on personal devices, TechCrunch reported. The lineup includes Nemotron 3.5 Super VL, a 120 billion parameter model for enterprise agents, and two smaller Bolt models for local use.
The details: Nvidia said Super VL matches the accuracy of the larger Nemotron 3 Ultra on instruction following and tool use. The company also said that when Super VL is post-trained on specific agentic tasks, such as document intelligence, financial accounting and security attack resilience, it completes them faster than competing models from Anthropic and OpenAI. Nvidia named Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce as users of Super VL.
On the device side, Nemotron 3.5 Bolt 74B-A9B 24GB is a 74 billion parameter model for local agentic work such as multistep reasoning, tool calling and web browsing. Bolt E4B is aimed at conversational agents. The Bolt models are designed for AI PCs with the Nvidia RTX Spark chip and for DGX Spark workstations.
What to watch: Super VL and Bolt 74B-A9B are available now, and Bolt E4B is due later this month. Nvidia's Kari Briski argued that organizations can post-train Super VL on their own data and own what they build. The speed comparisons against Anthropic and OpenAI models come from Nvidia and were not independently verified in the report, so teams should run their own evaluations before switching workloads.
Our take
Teams that want agents running on their own infrastructure or devices now have another open-weight option to test against closed models.