Nvidia Ships Open Models for Enterprise Agents and AI PCs

Nvidia launched a 120-billion-parameter open-weight model on Thursday, aiming squarely at companies that want to run AI agents on their own hardware. The model, Nemotron 3.5 Super VL, is built for always-on enterprise agents and edge computing, and Nvidia says it matches the instruction-following and tool-use accuracy of its larger Nemotron 3 Ultra while using fewer parameters.

The launch came with a direct claim from a named executive. "Any organization that wants to run agents on Nvidia infrastructure now has an open, optimized model for their tier without having to choose between capability and deployability," Kari Briski, Nvidia's vice president of Open Models and Software, wrote in the announcement blog post.

What does Super VL claim to beat?

Nvidia says Super VL completes tasks faster than competing models from Anthropic and OpenAI — the post names Opus 5, GPT 6.1 Sol, Luna and Astra — when post-trained on specific agentic workloads. Those workloads include document intelligence, financial accounting, scientific experimentation, medical insurance claims and security attack resilience.

Benchmark claims of this kind are routine at major model launches. Nvidia's counterweight is a customer list: Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce are all using Super VL, the company says.

Why open weights, and why now?

The strategic logic is straightforward. The fast-growing proprietary AI labs — companies that are simultaneously Nvidia's customers and the recipients of its investments — are diversifying their compute. They are turning to AWS's Trainium chips, to Cerebras, and to home-grown silicon of their own design. That erodes a reliable demand base for Nvidia's GPUs.

So Nvidia is leaning into a different trend: enterprises running their own open models inside their own data centers. "Unlike closed frontier models, organizations can post-train Super VL on their own data, deploy it on their own infrastructure and own what they build," Briski wrote.

This is not Nvidia's first move here. The company maintains a full family of Nemotron open-weight models covering multimodal AI, reasoning, coding, math and enterprise automation — models that compete directly with the proprietary frontier labs.

What are the Bolt models for PCs?

Alongside Super VL, Nvidia launched two lightweight Bolt models designed to run locally. One targets the new AI PCs built on the Nvidia RTX Spark chip; the other targets workstations running the DGX Spark. Partners including Microsoft and Dell are releasing these consumer- and knowledge-worker-focused machines into the market now.

Nemotron 3.5 Bolt 74B-A9B 24GB is the larger of the pair: a 74-billion-parameter model, trimmed to a 9-billion active-parameter design, built for local agentic work. Nvidia says it balances speed with reasoning to handle multistep reasoning, tool calling and web browsing. The smallest model, Bolt E4B, targets conversational agents.

The PC play addresses a real friction point. A new AI assistant seems to arrive daily from startups and big tech alike, but privacy-wary consumers hesitate to give cloud agents deep access to family, health and financial information. Nvidia's bet is that machines running models and agents locally will open a market for PCs powered by its chips. With local execution, PC users "can delegate everyday projects … without sending those workloads to the cloud," Briski wrote.

What is available today?

Super VL and Bolt 74B-A9B are available now. Bolt E4B arrives later this month, according to Nvidia.

The release tightens Nvidia's grip on both ends of the agent market: the data center, where it faces chip rivals, and the desktop, where it faces consumer distrust of cloud AI.