Nvidia Unveils Open-Weight Models for Enterprise Agents and AI PCs

Enterprise-Grade Nemotron 3.5 Super VL

Nvidia introduced the Nemotron 3.5 Super VL, a 120‑billion‑parameter model built for always‑on enterprise agents. Although it contains fewer parameters than the Nemotron 3 Ultra, Nvidia says it matches the Ultra’s accuracy on instruction‑following and tool‑use tasks.

In a blog post announcing the release, Kari Briski, vice president of Open Models and Software at Nvidia, wrote:

“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.”

The company also claims that after post‑training on specific agentic workloads, Super VL completes tasks faster than competing models from Anthropic and OpenAI, including Opus 5, GPT 6.1 Sol, Luna, and Astra. The targeted workloads cover document intelligence, financial accounting, scientific experimentation, medical insurance claims, and security attack resilience.

Several prominent firms have already adopted Super VL, according to Nvidia. These include Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce.

• Skydance
• EY
• Ramp
• Crowdstrike
• Perplexity
• Salesforce

Super VL joins Nvidia’s broader Nemotron family, which spans multimodal AI, reasoning, coding, mathematics, and enterprise automation. Unlike closed frontier models, organizations can post‑train Super VL on their own data, deploy it on their own infrastructure, and retain ownership of their creations, Briski explained in another excerpt:

“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.”

Lightweight Bolt Models for Local AI on PCs and Workstations

Nvidia also launched two lightweight Bolt models designed to run locally on new AI PCs powered by the Nvidia RTX Spark chip and on workstations using DGX Spark. These devices, aimed at consumers, programmers, and enterprise knowledge workers, are being released now through partners such as Microsoft and Dell.

The first Bolt offering, Nemotron 3.5 Bolt 74B‑A9B, is a 74‑billion‑parameter model packaged with 24 GB of memory. It is tuned for local agentic tasks, balancing speed with reasoning capability to handle multistep reasoning, tool calling, and web browsing.

The second, smaller Bolt model, Bolt E4B, targets conversational agents and is optimized for dialogue‑centric workloads.

Both Super VL and the Bolt 74B‑A9B are available immediately, while the Bolt E4B will follow later this month, Nvidia said.

Strategic Context and Market Positioning

While proprietary AI labs such as those at Anthropic and OpenAI are expanding their own chip ecosystems—adopting AWS Trainium, Cerebras, or building custom silicon—Nvidia is doubling down on open models that enterprises can run in their own data centers.

At the consumer level, Nvidia aims to accelerate the nascent market for local AI assistants. By keeping workloads on the device, users can delegate everyday projects without sending data to the cloud, Briski noted:

“PC users can delegate everyday projects … without sending those workloads to the cloud.”

Privacy concerns surrounding deep personal data sharing have slowed broader adoption of cloud‑based agents. Nvidia’s approach with the Bolt models offers a path for users to retain control over their information while still benefiting from AI assistance.

Overall, the new open‑weight models reflect Nvidia’s strategy to provide enterprises and end‑users with flexible, capable AI tools that can be deployed on Nvidia hardware ranging from data‑center GPUs to consumer‑grade AI PCs.