News in short

• Nvidia has introduced new open-weight Nemotron 3.5 models for enterprise AI agents and local computing.
• Nemotron 3.5 Super VL is a 120-billion-parameter model designed for enterprise agent workloads.
• The smaller Bolt models target local AI tasks on Nvidia RTX Spark PCs and DGX Spark workstations.
• Nvidia says businesses can customise Super VL using their own data and deploy it on their own infrastructure.
• Super VL and Bolt 74B-A9B are available, while Bolt E4B is expected later in October 2026.

Nvidia has introduced new open-weight artificial intelligence models designed to run enterprise agents and AI assistants locally on computers. The latest additions to its Nemotron family target two growing use cases: business automation and AI workloads that do not need to rely entirely on cloud services.

The lineup includes Nemotron 3.5 Super VL, a 120-billion-parameter model for enterprise agents, and two smaller Bolt models designed for local computing. Nvidia says the models aim to balance AI capability with deployment flexibility.

The announcement also reflects the company’s push to support businesses that want greater control over their AI systems, data and infrastructure.

Nemotron 3.5 Super VL targets enterprise AI agents

Nvidia’s Nemotron 3.5 Super VL is designed for enterprise applications that require agents to follow instructions, use tools and complete multistep tasks.

Although it has fewer parameters than Nvidia’s Nemotron 3 Ultra, the company says Super VL can deliver similar accuracy in instruction-following and tool-use tasks.

Nvidia also claims the model can complete specialised agent tasks faster than competing models from Anthropic and OpenAI after post-training for specific workloads. These tasks include document intelligence, financial accounting, scientific experimentation, medical insurance claims and security attack resilience.

Such performance claims depend on the tasks, evaluation methods and model configurations involved. Independent testing would help establish how the model compares across broader real-world workloads.

Nvidia says companies including Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce are using Super VL. The company has not detailed the extent of each organisation’s deployment in the supplied announcement.

Bolt models bring AI agents to local PCs

Nvidia has also introduced two smaller models under the Nemotron 3.5 Bolt family. These target local AI workloads on Nvidia RTX Spark PCs and DGX Spark workstations.

The Nemotron 3.5 Bolt 74B-A9B 24GB model contains 74 billion parameters. Nvidia designed it for multistep reasoning, tool calling and web browsing, with a focus on balancing speed and reasoning capabilities.

The smaller Nemotron 3.5 Bolt E4B is aimed at conversational agents. Nvidia says this model will arrive later in October 2026.

The local deployment approach could let users run certain AI tasks directly on compatible hardware instead of sending every request to remote servers. However, performance will depend on the device, memory capacity, workload and model configuration.

Nvidia is positioning these models alongside its RTX Spark computing hardware, which targets consumers, programmers and enterprise knowledge workers.

Why Nvidia is pushing open-weight AI

Open-weight models give organisations access to model weights, allowing them to customise and deploy models within supported licensing and technical constraints. This differs from relying solely on proprietary AI services accessed through a provider’s platform.

Nvidia says businesses can post-train Super VL on their own data and run it on their own infrastructure. This approach could give companies more control over deployment, integration and sensitive workloads.

Local AI assistants also offer a potential privacy advantage. Users may prefer to process certain documents and personal tasks on their own computers rather than send all associated data to cloud services. Local execution does not automatically guarantee privacy, however; security still depends on software, permissions and how the system handles information.

The strategy also supports Nvidia’s broader hardware business. By providing models optimised for its infrastructure, the company can encourage organisations to build AI workloads around its GPUs and systems.

Availability of Nvidia Nemotron 3.5 models

Nvidia says Nemotron 3.5 Super VL and Bolt 74B-A9B are available now. Bolt E4B is expected later in October.

The models expand Nvidia’s existing Nemotron portfolio, which covers reasoning, coding, mathematics, multimodal AI and enterprise automation.

As businesses explore AI agents that can carry out tasks rather than simply answer questions, the demand for models that offer flexibility, customisation and local deployment could grow. Nvidia’s latest releases target that demand, although their practical value will depend on performance, deployment costs and the reliability of the resulting agents.