NVIDIA Nemotron 3.5 Super VL and Bolt models let enterprises run agentic AI on local hardware
New open-weight models target enterprise agents and local hardware, offering efficiency gains across the deployment spectrum.
Key points
- Nemotron 3.5 Super VL is a 120-billion-parameter model designed for high-accuracy agentic tasks.
- Two Bolt variants enable local execution on edge devices with reduced memory requirements.
- Organizations can post-train these models on proprietary data for specialized workflows.
NVIDIA has released Nemotron 3.5 Super VL alongside two new Bolt variants, expanding its open model family to cover deployments from data centres to local hardware. As TechSec Observer reported on 2026-10-16, the 120-billion parameter model targets background agent workflows via compact Bolt variants.
Root cause
The release aims to provide open, optimized models for every deployment tier. Super VL delivers the fastest task completion and highest accuracy for specialized always-on agents. It achieves leading accuracy on multimodal agentic tasks comparable with NVIDIA’s largest reasoning model at roughly one-fifth the parameter count.
Attack path
This is a software release, not a security incident. There is no attack path to analyse. The focus is on how organisations integrate these models into existing stacks. Super VL is built for instruction following and tool use, enabling agents to execute multistep business workflows.
Affected versions
The release includes three distinct models. Super VL is available now for enterprise servers. Bolt 74B-A9B 24GB is also available, optimized for local agentic workloads using 2-bit quantization. Bolt E4B 4GB, designed for conversational agents, becomes available on 23 October.
Mitigation
No mitigation is required as this is a product launch. However, teams should assess whether their current infrastructure supports the memory requirements of the Bolt variants. Super VL and Bolt 74B-A9B are ready for immediate deployment on compatible NVIDIA hardware.
What it means for organisations
Operations teams must evaluate whether local execution reduces cloud dependency costs. Security leads should verify that post-training processes align with data governance policies, especially when using proprietary datasets. Procurement should review vendor lock-in risks associated with specific hardware requirements. Engineering managers need to determine if the speed gains justify migrating existing agent workloads to these new open-weight options.
What to do and how to stay safe: NVIDIA
- Audit current agent workloads to identify candidates for local deployment using Bolt variants.
- Verify hardware compatibility for 24GB and 4GB memory constraints before deploying edge models.
- Review data privacy policies before post-training Super VL on sensitive internal datasets.
- Monitor vendor documentation for updates on Bolt E4B availability in late October.
General security guidance from the TechSec Observer newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is the parameter count of Nemotron 3.5 Super VL?
It is a 120-billion-parameter model designed for high-accuracy agentic tasks.
When is the Bolt E4B 4GB model available?
It is scheduled for release on 23 October.
Can these models run on local hardware?
Yes, the Bolt variants are optimized for local execution on devices like RTX and Jetson.
