Key points
- Nvidia released Nemotron 3.5 Super VL, a 120-billion-parameter model for agentic work.
- Two new edge models, Nemotron 3.5 Bolt, target local and conversational agent use cases.
- The models aim to provide faster task completion and higher accuracy for specialized agents.
Nvidia has made three new open-weight AI models available to developers and enterprises. The primary release is Nemotron 3.5 Super VL, a 120-billion-parameter model designed for agentic workflows. Nvidia states this model delivers the fastest task completion and highest accuracy for specialized, always-on agents. It is positioned as the fastest open agentic model in its class, operating five times faster than the next best competitor, according to Artificial Analysis benchmark testing.
In plain English
These models are intended for tasks that require automated agents to complete high-volume work over days or weeks. Nemotron 3.5 Super VL focuses on speed and accuracy for these long-running processes. The company also released two edge models: Nemotron 3.5 Bolt 74B-A9B 24GB and Nemotron 3.5 Bolt E4B 4GB. Edge AI allows models to operate locally and off-grid. The larger Bolt model is built for local agentic use, while the smaller version targets fast, conversational agents. Both edge models deliver the highest intelligence per gigabyte of memory relative to Nvidia’s other offerings.
The background
Open-weight models are seeing increased adoption following the Hugging Face hack, according to ZDNET. US labs are releasing more open models to challenge China-based AI labs, which have held the upper hand in competitive open-weight AI for years. Nvidia’s release is part of this surge in Western-made open models. The advantage of open models is that businesses can post-train them on their own data and run them on their own infrastructure, offering greater personalization and security compared to proprietary models.
What changes now
Enterprises can now tailor Nemotron 3.5 Super VL to their specific needs through post-training. Nvidia cited a 44% gain in medical insurance claims accuracy as an example of how post-training can enhance capabilities. While the base model did not top proprietary frontier models on its own, it allows for highly specific gains when customized. The base model is available today, enabling immediate integration into workflows that require specialized, always-on agents.
Who needs to act first
Teams evaluating these models should prioritize testing the Nemotron 3.5 Super VL in isolated environments before deploying to production. Verify that the model meets specific accuracy requirements for your use case through benchmarking. Ensure that infrastructure can support the 120-billion-parameter model’s resource needs. For edge deployments, assess whether the Bolt models fit within existing memory constraints. Monitor performance closely during initial rollout to identify any deviations from expected behavior.
Background: Edge computing
Edge computing processes data near its source rather than a central cloud. This reduces latency but fragments security management. You must secure each node individually, manage firmware updates remotely, and ensure data integrity during transmission to local processing units.
Read the full guide: Edge Computing FAQs: Security Risks, Architecture, and Real-World Answers
What to do and how to stay safe: Nvidia
- Review the model documentation to understand the specific requirements for Nemotron 3.5 Super VL and the Bolt variants.
- Test the models in a controlled environment to verify accuracy and performance for your specific agentic tasks.
- Ensure your infrastructure has sufficient memory and compute resources to run the 120-billion-parameter model or the edge variants.
- Monitor model outputs during initial deployment to confirm they meet your organization’s accuracy and security standards.
Step-by-step guide: How Passkeys Work: The Cryptographic Mechanics Behind Passwordless Login
General security guidance from the Patch Review newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is Nemotron 3.5 Super VL?
It is a 120-billion-parameter open-weight AI model released by Nvidia for agentic workflows, aiming for fast task completion and high accuracy.
Are there edge versions of these models?
Yes, Nvidia released Nemotron 3.5 Bolt 74B-A9B 24GB for local agentic use and Nemotron 3.5 Bolt E4B 4GB for fast, conversational agents.
How does Nemotron 3.5 Super VL compare to other models?
According to Artificial Analysis, it is five times faster than the next best open agentic model, outperforming InclusionAI’s Ling 3.0 Flash and DeepSeek’s V4.1 Flash.
