Nvidia has expanded its open-weight artificial intelligence portfolio with Nemotron 3.5 models designed for enterprise AI agents, automated business workflows and local computing on compatible systems.
Announced on October 15, 2026, the lineup includes Nemotron 3.5 Super VL, a model with 120 billion total parameters, and two smaller models designed for local AI applications.
Nvidia says the releases aim to help developers build specialised AI agents that can reason, use tools and complete multistep tasks across different computing environments.
According to Nvidia’s official announcement, Super VL and the Bolt 74B-A9B 24GB model are available, while Bolt E4B is scheduled for October 23.
The release extends Nvidia’s strategy beyond AI hardware by giving developers additional models they can customise and deploy on compatible infrastructure.
Nemotron 3.5 Super VL Targets Enterprise AI Workloads
Nemotron 3.5 Super VL is designed for businesses that need AI systems to process documents, follow instructions, call external tools and carry out complex workflows.
The model has 120 billion total parameters, with 12 billion active parameters in its mixture-of-experts architecture. Nvidia’s official model card identifies it as a vision-language model supporting text, images and video.
Nvidia says Super VL can be adapted through specialised post-training for tasks such as document intelligence, financial accounting, scientific research and medical insurance claims.
The company also reports strong performance and throughput on selected agentic benchmarks. These are Nvidia’s own performance claims and should be assessed against the published evaluation methods, hardware configurations and independent testing before being generalised to all business applications.
Nvidia has named Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce among organisations using Super VL, according to its announcement.
For enterprises, the model offers another option for developing specialised AI systems using their own data and deployment infrastructure, rather than depending entirely on closed, remotely hosted AI services.
Smaller Bolt Models Bring AI Agents to Local Systems
Nvidia has also introduced two smaller models aimed at bringing agentic AI capabilities to local computers and workstations.
The first, Nemotron 3.5 Bolt 74B-A9B 24GB, has 74 billion total parameters and is designed to fit within a 24GB memory budget through aggressive quantisation. Its architecture activates a smaller portion of the model’s parameters for each input.
According to Nvidia’s technical report, the model targets everyday tasks such as web research, tool use, coding and other multistep workflows. It is intended to run on compatible Nvidia RTX systems and DGX Spark hardware.
The second model, Nemotron 3.5 Bolt E4B 4GB, is designed for lighter conversational and interactive AI applications. Nvidia says it is intended for use cases including voice assistants and embedded interfaces.
At the time of the announcement, the E4B model was scheduled for release on October 23.
Local AI Could Give Users More Control Over Data
Running an AI model locally can reduce the need to send every prompt or document to a cloud service. This may appeal to organisations handling sensitive financial records, private business documents or personal information.
Local deployment can also give developers greater control over model configuration, integration and data-processing workflows.
However, running a model on a personal computer does not automatically guarantee privacy or security. Connected applications, network requests, telemetry and third-party services may still transmit information externally. Users and organisations must assess the complete system, not just the model itself.
Hardware requirements also matter. A model’s total parameter count does not, by itself, determine its memory usage or speed. Quantisation, active parameters, software optimisation and available hardware all affect real-world performance.
Nvidia’s Nemotron developer portal provides additional information about its model family and supported development frameworks.
Nvidia Expands Its Open-Weight AI Strategy
The Nemotron 3.5 releases reflect growing interest in models that developers can adapt for specific tasks and deploy across cloud, data-centre and local environments.
Open-weight models make model parameters available for use, but the term does not automatically mean that every component, including training data and development procedures, is fully open. Businesses must also review the applicable licence before commercial deployment.
Nvidia provides model documentation, technical reports and deployment resources through its official developer ecosystem. Developers can use these resources to assess capabilities, hardware compatibility and customisation options before committing to a production system.
The wider AI market is also moving towards a mix of cloud-based services, specialised chips and locally deployed models. Nvidia’s latest release gives developers more options within its own hardware and software ecosystem.
The main question now is how the models perform in practical deployments, particularly in reliability, inference speed, operating costs and the handling of sensitive information.
Independent testing and real-world enterprise results will help determine where Nemotron 3.5 offers a measurable advantage.
