Nvidia launches Nemotron 3.5 Super VL and Bolt models for enterprise agents

Nvidia launched new open-weight artificial intelligence models on Thursday, targeting both enterprise data centers and local device computing. The release includes the Nemotron 3.5 Super VL for large-scale operations and two lightweight Bolt models for edge devices, according to a report by TechCrunch.

The Nemotron 3.5 Super VL contains 120 billion parameters and is designed for always-on agents within enterprise environments. Although it uses fewer parameters than the previous Nemotron 3 Ultra, Nvidia states it delivers similar accuracy in instruction-following and tool-use scenarios. This model aims to provide organizations with optimized performance without sacrificing deployability, according to Kari Briski, Nvidia’s vice president of Open Models and Software.

Technical Capabilities and Competitors

Nvidia asserts that the Super VL model completes specific agentic tasks faster than competing proprietary models. These competitors include Anthropic’s Opus 5, OpenAI’s GPT 6.1 Sol, Luna, and Astra, according to the company’s announcement. The benchmarked tasks cover document intelligence, financial accounting, scientific experimentation, medical insurance claims, and security attack resilience.

While performance claims are common in model announcements, Nvidia lists several major companies using the Super VL model to validate its utility. These organizations include Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce, according to the source material. This list highlights the model’s adoption across diverse sectors, from media to cybersecurity and financial services.

Hardware and Deployment Strategy

At the consumer and workstation end, Nvidia introduced two lightweight Bolt models intended for local execution. These models run on new AI PCs equipped with the Nvidia RTX Spark chip and workstations using the DGX Spark hardware. Partners such as Microsoft and Dell are releasing these devices to consumers, programmers, and enterprise knowledge workers immediately, according to TechCrunch.

Nvidia is emphasizing the ability for organizations to post-train the Super VL model on their own data. This approach allows companies to deploy the model on their internal infrastructure and retain ownership of their builds, Briski wrote. This strategy contrasts with closed frontier models, offering greater control over data privacy and model customization for enterprise users.

What should you do?

Security and technology professionals should evaluate whether local model execution aligns with their data governance policies. Organizations considering these models should verify the specific hardware requirements for the RTX Spark and DGX Spark chips. Teams should also assess the post-training capabilities to ensure they match internal data handling standards.

What to do and how to stay safe: Nvidia

• Review internal data governance policies to determine if local model execution meets privacy requirements.
• Verify hardware compatibility for RTX Spark and DGX Spark chips before deploying Bolt models.
• Assess the post-training capabilities of open-weight models against internal data handling standards.
• Monitor performance benchmarks for agentic tasks like document intelligence and security resilience.

Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Which companies are using Nvidia's new Nemotron 3.5 Super VL model?

Nvidia lists Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce as users of the Super VL model.

What hardware runs the new Bolt models?

The Bolt models run on AI PCs with the Nvidia RTX Spark chip and workstations with the DGX Spark.

How does Nemotron 3.5 Super VL compare to previous models?

It has 120 billion parameters and delivers similar accuracy to the Nemotron 3 Ultra in instruction-following tasks.