# Nvidia launches open Nemotron 3.5 models for enterprise and PC agents

published: 2026-10-19T16:05:23.338Z
topic: infrastructure
source: https://techcrunch.com/2026/10/19/nvidia-launches-new-open-models-to-run-agents-in-the-enterprise-and-on-pcs/
url: https://aiagentnews.io/stories/nvidia-launches-open-nemotron-3-5-models-for-enterprise-and-pc-agents

Nvidia has released Nemotron 3.5 Super VL, a 120-billion-parameter open-weight model for enterprise agents, plus two smaller Bolt models for local use on AI PCs and DGX Spark workstations. Super VL and Bolt 74B-A9B are available now, and Bolt E4B is due later this month.

Nemotron 3.5 Super VL is aimed at enterprise agents. Nvidia says it uses fewer parameters than Nemotron 3 Ultra but delivers similar accuracy on instruction-following and tool-use tasks. The company also claims it finishes post-trained agentic tasks faster than competing models from Anthropic and OpenAI, and lists Salesforce, EY, Ramp, CrowdStrike and Perplexity among companies using it.

The two Bolt models are built to run on-device. Nemotron 3.5 Bolt 74B-A9B 24GB, with 74 billion parameters, is meant for multistep reasoning, tool calling and web browsing on local machines. The smaller Bolt E4B targets conversational agents and is due later this month.

Nvidia is pitching the models to PCs powered by its RTX Spark chip and to DGX Spark workstations, which partners such as Microsoft and Dell are shipping now. The company says these machines let users run agent workloads without sending them to the cloud.

## Why it matters if you build agents

Teams that want to self-host agents now have another open-weight option from Nvidia, with Super VL designed to be post-trained on their own data and deployed on their own infrastructure.

Developers targeting Nvidia PCs or DGX Spark machines can test local agents with Bolt 74B-A9B today, though Bolt E4B is not yet available.

Nvidia's benchmark claims come from its own post-training tests, so teams should check speed and accuracy on their own agent workloads before switching models.