Nvidia Launches Open Nemotron 3.5 Models for Enterprise Agents and Local PCs

Nvidia has launched a new suite of open-weight AI models, directly targeting two growing trends: enterprise companies running their own AI in private data centers, and consumers using AI assistants that run locally on their PCs. The announcement, made on October 19, 2026, and reported by TechCrunch, signals a strategic push to provide optimized models for every tier of deployment, from massive data centers to personal workstations.

What new models did Nvidia announce?

Nvidia announced three new models under its Nemotron family. The flagship is the Nemotron 3.5 Super VL, a 120-billion-parameter model designed for complex, agentic tasks in enterprise settings. According to Nvidia, it matches the accuracy of larger models in instruction-following and tool-use but is optimized for faster task completion. At the other end are two lightweight "Bolt" models for local execution. The Nemotron 3.5 Bolt 74B-A9B is a 74-billion-parameter model for multistep reasoning and tool calling on PCs, while the smaller Bolt E4B is aimed at conversational agents.

Why is Nvidia focusing on open models for enterprises?

The move is a direct response to a clear market demand. As reported by TechCrunch, Nvidia is leaning into the trend of enterprises wanting to run their own open models on their own infrastructure. Kari Briski, Nvidia's vice president of Open Models and Software, stated in the announcement blog post that unlike closed models, organizations can post-train the Super VL on proprietary data, deploy it on their own systems, and retain ownership of what they build. This offers control, privacy, and customization that closed API models cannot. Early adopters named by Nvidia include Skydance, EY, and Salesforce.

How do the new Bolt models change the AI PC landscape?

The Bolt models are designed to run on new AI PCs equipped with Nvidia's RTX Spark chip and workstations with the DGX Spark. Partners like Microsoft and Dell are bringing these machines to market now. The key proposition is privacy and latency: PC users can delegate everyday projects to a local AI agent without sending sensitive data—about health, finances, or family—to the cloud. This addresses growing consumer privacy concerns while attempting to create a new market for Nvidia-powered AI PCs.

What does this mean for AI developers and creators?

For developers building agentic applications, the new models offer more deployment options. The Super VL provides a powerful, open base for post-training on domain-specific enterprise tasks like document intelligence or scientific experimentation. The availability of the Bolt 74B model for local PCs opens the door to developing responsive, private desktop assistants that can perform web browsing and multistep reasoning without an internet connection. This expands the architectural possibilities for AI applications beyond cloud-only dependencies.

How can you explore agentic AI with Grafy?

While Grafy does not host the new Nemotron models directly, the principles of autonomous AI agents are central to platforms like Grafy. You can experiment with agentic workflows using Grafy's Operator (/operator), a browsing agent designed to perform multi-step web tasks. For a deeper technical understanding, use Grafy Chat (/chat) to query multiple AI models about agent architectures, or load research papers on the topic into Reader (/reader) for analysis. The release underscores a broader shift toward specialized, deployable AI that Grafy's multi-model, multi-app workspace is built to accommodate.

Frequently asked questions

What are "open-weight" models?

Open-weight models have their architecture and trained parameters publicly released. This allows organizations to download, run, and modify them on their own hardware without relying on a vendor's API, enabling greater control, customization, and data privacy compared to closed models like GPT-4.

When are the Nvidia Bolt models available?

According to Nvidia, the Super VL and Bolt 74B-A9B models are available now. The smallest conversational model, Bolt E4B, is scheduled for release later in October 2026. They are designed to run on new AI PCs with Nvidia's RTX Spark chip.

How does this affect competition with OpenAI and Anthropic?

Nvidia's move positions its open models as direct alternatives to proprietary frontier models from companies like OpenAI and Anthropic, which are also its customers. By offering enterprises an open, optimizable option for deployment on Nvidia infrastructure, it seeks to capture more of the AI stack value as these labs also explore other chip suppliers.

---

Source: TechCrunch — AI — Nvidia launches new open models to run agents in the enterprise and on PCs