NVIDIA has launched Nemotron 3.5 Super VL, an open-weight artificial intelligence model designed to power specialized AI agents that can understand visual information, reason through complex tasks and use external tools. Announced on October 15, 2026, the model combines text, image, document and video understanding with agentic capabilities in a single system.
According to NVIDIA, Nemotron 3.5 Super VL can deliver up to five times higher throughput than comparable open models in its size category on agentic tasks. The company says the model is designed to help businesses build AI agents capable of handling complex, multistep workflows more efficiently.
NVIDIA Nemotron 3.5 Super VL brings multimodal reasoning
The model uses a mixture-of-experts architecture containing 120 billion parameters, with 12 billion parameters activated per token. This design aims to balance model capacity with inference efficiency.
Unlike systems that rely on separate models for text processing and visual analysis, Nemotron 3.5 Super VL combines these capabilities within one model. It can interpret documents, charts, images and video while following instructions, reasoning through tasks and calling tools.
These capabilities could support AI agents that research information across websites, analyse financial documents, review code and identify potential cybersecurity threats. The model is also designed to extract structured information from tables, charts and complex document layouts.
Businesses can customise AI agents
NVIDIA is positioning Nemotron 3.5 Super VL as a foundation for specialised enterprise AI applications. Developers can adapt the model using their own data, tools, operating procedures and policies to suit particular business requirements.
The company highlighted examples from organisations working on specialised AI workflows. Scale AI has worked on automating vendor credit processing, while financial operations platform Ramp is using the model to convert receipt images into structured data. AI21 has also adapted the model for threat intelligence and vulnerability analysis.
NVIDIA reports substantial efficiency and performance improvements in these customised applications. However, the results vary by workload and are based on the respective organisations’ evaluations rather than a universal performance guarantee.
Open model gives developers greater control
Nemotron 3.5 Super VL is available with model weights, datasets and training resources that developers can use to customise their applications. NVIDIA says teams can use its NeMo tools for supervised fine-tuning, low-rank adaptation and reinforcement learning.
The model can be deployed in supported local, data-centre and cloud environments, giving organisations options for managing their infrastructure and data requirements. NVIDIA also supports deployment through its NVIDIA NIM platform and inference optimisation through NVIDIA Dynamo.
The company says its inference infrastructure can improve serving capacity and reduce latency, depending on the deployment configuration.
What the launch means for AI agents
The launch reflects the growing focus on AI agents that can do more than generate text or answer questions. By combining visual understanding, reasoning and tool use, Nemotron 3.5 is designed to handle workflows that involve interpreting information, making decisions and carrying out multiple steps.
For businesses, the potential benefits include document processing, research automation, software development and security analysis. Actual results will depend on the task, hardware, model customisation and deployment setup.
With Nemotron 3.5 Super VL, NVIDIA is expanding its open-model ecosystem for organisations seeking greater control over how specialised AI agents are trained and deployed.
Read more : NVIDIA is also expanding industrial AI adoption through its Germany–NVIDIA AI deal.
