Nvidia introduces new open-weight models for enterprise AI agents

Nvidia has introduced new open-weight models designed to support always-on agents in enterprise settings and edge computing applications. The Nemotron 3.5 Super VL is a large model boasting 120 billion parameters, making it suitable for use by enterprises that require high-performance from their agents.

The Nemotron 3.5 Super VL model achieves similar accuracy levels as its larger counterpart, the Nemotron 3 Ultra, in tasks such as instruction-following and tool-use. This efficiency comes despite having fewer parameters than the Nemotron 3 Ultra, indicating that Nvidia has optimized the model for specific use cases.

Nvidia also launched two lightweight models called Bolt, which are designed to run locally on AI PCs (powered by the Nvidia RTX Spark chip) or workstations equipped with the DGX Spark. These PCs, intended for consumers and programmers, as well as enterprise knowledge workers, are being released through partners such as Microsoft and Dell.

The new open models have been touted by Nvidia as providing organizations with a straightforward solution for deploying agents without having to choose between model performance and deployability.

Nvidia's new Super VL models have been adopted by a number of prominent companies, including Skydance, EY, Ramp, Crowdstrike, Perplexity and Salesforce. These organizations are reportedly using the open models to deploy AI agents in their operations.

The release of Super VL marks the latest development in Nvidia's efforts to provide open-weight models that compete with proprietary offerings from companies like Frontier Labs. The GPU giant has previously released a family of Nemotron models that support various tasks, including multimodal AI and enterprise automation.

However, some of these proprietary labs are now looking beyond Nvidia for their AI needs, turning to alternative chips and system makers such as AWS's Trainium and Cerebras, or even developing their own custom hardware. This trend has prompted Nvidia to focus on the growing demand from enterprises for open models that can be deployed in-house.

Organizations using Super VL models can allegedly post-train them on their own data, deploy them on their own infrastructure, and retain ownership of what they build. This approach is said to offer a more flexible alternative to closed frontier models.

The introduction of smaller PC models also signals Nvidia's push into the consumer AI assistant market. As new AI-powered assistants emerge from startups and established tech companies alike, some consumers are becoming increasingly wary about granting access to sensitive information such as family details, health records, and financial data.

Nvidia's new open models aim to enable local running of agents on PCs, allowing users to delegate everyday tasks without sending workloads to the cloud.

The two Bolt models, Nemotron 3.5 Bolt 74B-A9B 24GB and Bolt E4B, are designed for specific use cases. The larger model is a 74-billion parameter design intended to balance speed with reasoning capabilities, making it suitable for tasks that involve multistep reasoning, tool calling, and web browsing.

Super VL and Bolt 74B-A9B are now available, while the smaller Bolt E4B will be released later this month, according to Nvidia's announcement.

The recent announcement from Nvidia has marked an important milestone in the development of open models that can run agents in various environments, including enterprise settings and personal computers. These new models are designed to provide more flexibility and scalability for users who need to deploy AI-powered agents in their workflows.

Super VL and Bolt 74B-A9B have been made available immediately, with the smaller Bolt E4B set to be released later this month according to Nvidia's schedule. This move is expected to further accelerate the adoption of open models in various industries and applications. The availability of these new models will undoubtedly provide users with more options for integrating AI-powered agents into their systems.