NVIDIA Expands Open Model Portfolio With Nemotron 3.5 Super VL and Kumo

NVIDIA announced an expansion of its open model portfolio, introducing new tools for language, robotics, biology and quantum computing. The release includes the Nemotron 3.5 Super VL and Nemotron 3.5 Bolt models, designed to help developers build applications tailored to local languages and specific industry needs. This move supports organizations seeking to customize AI with local data and deploy it on infrastructure they control.

The new Nemotron Retriever models handle embedding, reranking and document parsing tasks. These additions join the broader Nemotron family, which provides open language models, datasets and recipes for building agents. The models support both text and images, allowing agents to orchestrate workflows and reason through complex problems at a lower cost. NVIDIA states these tools give enterprises a foundation to build applications based on their own knowledge.

Industry Partners and Ecosystem

Nemotron models incorporate data and expertise from the NVIDIA Nemotron Coalition, which includes European partners Black Forest Labs, H-Company and Mistral. Major enterprises such as Dassault Systèmes, SAP, Salesforce, Shopify and WPP are working with these models to create custom solutions. Professional services providers like Accenture and EY are also utilizing the datasets to build domain-specific agents for their clients.

A wide ecosystem of post-training partners assists enterprises in customizing Nemotron for specific workloads. This group includes AfterQuery, AgileRL, AI21, Google Cloud, Mercor, Nebius, Scale AI and Trajectory. These partners help organizations fine-tune the models for specialized tasks, reducing the complexity of developing proprietary AI systems from scratch.

Structured Data and Robotics Tools

NVIDIA also launched the Kumo family of open weight foundation models and libraries for structured business data. Part of the NVIDIA Agent Knowledge Foundation, Kumo helps agents forecast outcomes, score fraud risk and recommend actions. It reduces the cost and complexity of building separate machine learning pipelines for every task by shortening the time to prediction.

The Kumo portfolio includes NVIDIA Kumo Structured for relational and tabular prediction, forecasting and anomaly detection. It also features NVIDIA Kumo Engine, an open library stack for developing and deploying structured data foundation models. Early adopters include Cognite, Coinbase, Databricks, DoorDash, Revolut and Tubi. Snowflake customers are already using these models in advertising, energy and financial services sectors.

Robotics and Autonomous Systems

NVIDIA Cosmos 3.5 world models accelerate robotics training and simulation. The family of open world foundation models helps developers build robots, autonomous vehicles and vision agents that understand the physical world. New post-training and optimization tools allow developers to improve world models using less compute power.

Simulation checkpoints enable evaluation of behavior in interactive driving and robotic manipulation scenarios before real-world testing. NVIDIA also introduced Cosmos Data Factory, which comprises blueprints and core agent skills. This tool helps developers curate, label, generate and evaluate training data for driving scenarios, product defects and image augmentation.

What should you do?

Security and technology professionals should assess whether their current AI development pipelines can integrate with these new open models. Teams should review data governance policies to ensure local data customization complies with regional regulations. Organizations should also evaluate the compute requirements for fine-tuning these models on their own infrastructure.

What to do and how to stay safe: NVIDIA

• Review internal data governance policies to ensure compliance when customizing open models with proprietary or local data.
• Evaluate current infrastructure capabilities to determine if they can support the compute requirements for fine-tuning these new models.
• Assess existing machine learning pipelines to identify opportunities where Kumo or Nemotron models could reduce complexity or cost.
• Monitor vendor documentation for updates on security best practices when deploying open weight models in production environments.

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 European partners are involved in the Nemotron Coalition?

The NVIDIA Nemotron Coalition includes European partners Black Forest Labs, H-Company and Mistral, according to the announcement.

What industries are using the new Kumo models?

Early adopters include financial services, advertising, energy and retail, with companies like Coinbase, Revolut and Databricks working with the models.

Does NVIDIA provide tools for robotics simulation?

Yes, NVIDIA introduced Cosmos Data Factory and simulation checkpoints for evaluating behavior in interactive driving and robotic manipulation scenarios.