Nvidia Nemotron 3.5 Super VL: Open AI Agent Model for Enterprise

Introduction: An AI Worker That Never Clocks Out

Imagine a digital employee that quietly reads documents, looks at images, watches videos and works through business tasks around the clock. That is the idea behind Nemotron 3.5 Super VL, the new open multimodal model from Nvidia, built for enterprise agents that run continuously in the background. Two compact Bolt models arrive alongside it to bring agents to local PCs and edge devices.

Key Facts

• Super VL has 120 billion parameters and handles text, images, documents and videos.
• Nvidia developed it with the Nemotron Coalition, a group that includes Mistral AI, Thinking Machines Lab, Perplexity and Black Forest Labs.
• Nemotron 3.5 Bolt 74B-A9B 24GB runs agent tasks on local machines. The smaller Bolt E4B 4GB targets fast voice assistants and will follow later this month.
• Dozens of organizations are reportedly already fine-tuning Super VL.

Both models offer the “highest intelligence per gigabyte of memory,” Nvidia claims.

Technical Background

Mixture-of-Experts: Calling Only the Specialists You Need

Super VL uses a Mixture-of-Experts (MoE) architecture. Think of a large hospital: not every doctor sees every patient, only the relevant specialists do. Here, only 12 billion of the 120 billion parameters are active per token, which keeps the model fast and cheaper to run.

Distilled From a Bigger Sibling

Nvidia distilled Super VL from Nemotron 3 Ultra, its largest reasoning model. On instruction following, tool calling, multi-step business processes and image analysis, it is supposed to match Ultra’s accuracy with about one-fifth the parameters. Nvidia says it scores higher than Ultra in agent tests from Hermes and LangChain, and beats its predecessor Nemotron 3 Super on the SWE-bench and Terminal-Bench coding benchmarks.

Speed and Compression

According to Nvidia, citing the Artificial Analysis leaderboard, Super VL delivers up to five times the throughput of similarly sized open models on a single GPU (Graphics Processing Unit) for agent tasks. In an independent test by EXO Labs on the DGX Spark, no other model was both more accurate and faster. The model is also compressed with Nvidia’s 4-bit NVFP4 format, and on Blackwell GPUs it runs entirely in 4-bit precision.

Global Implications

Nvidia is pitching Super VL against closed frontier models for companies that want to fine-tune on their own data and run it on their own servers, free from a provider’s terms of service. Examples it gives include an insurer screening claims for fraud or a financial firm scoring creditworthiness by its own criteria. Customization tools include the NeMo framework, Python pipelines and open datasets.

There is a catch worth noting: the model is open, but getting the most out of it means buying Nvidia hardware, and every Nemotron variant is optimized for it. Openness here also strengthens Nvidia’s hardware ecosystem. For businesses in regulated or privacy-sensitive sectors worldwide, the ability to keep data in-house may be attractive, though real-world results beyond Nvidia’s own claims remain to be seen.

Conclusion and Outlook

Nemotron 3.5 Super VL signals a shift toward efficient, always-on agents rather than ever-bigger chatbots. Watch for the Bolt E4B release later this month, independent benchmarks, and how many enterprises move from fine-tuning experiments to production.

---

Stock Market Impact Analysis

Publicly traded companies directly or indirectly affected by this news. Always conduct independent research before making investment decisions.

Investor Impact by Stock

※ Price data via yfinance (may include after-hours). Retrieved: 2026-10-16 00:03 UTC

---

🛒 Recommended Gear

• The Agentic AI Bible — Building Goal-Driven LLM Agents
• Build a Reasoning Model From Scratch (Sebastian Raschka)

As an Amazon Associate, this site earns from qualifying purchases.

---

Sources (1 articles)

• [The Decoder] Nvidia’s Nemotron 3.5 Super VL is an open model built to run enterprise AI agents nonstop

※ This article synthesizes and analyzes the above sources. Generated: 2026-10-16 00:03