Nvidia has released new open-weight models to power agents in enterprise environments and on local PCs.
Two new tools for different jobs
The Nemotron 3.5 Super VL is a large model containing 120 billion parameters. It targets enterprise use cases. Nvidia states it matches the accuracy of its larger Nemotron 3 Ultra model in instruction-following and tool-use tasks while using fewer parameters.
At the other end of the scale, the company launched two lightweight Bolt models. These run locally on new AI PCs equipped with the Nvidia RTX Spark chip and workstations using the DGX Spark. Microsoft and Dell are distributing these machines to consumers, programmers and knowledge workers.
Kari Briski, Nvidia’s vice president of Open Models and Software, wrote in the announcement that any organisation wanting to run agents on Nvidia infrastructure now has an open, optimised model for their tier without having to choose between capability and deployability.
Nvidia claims the Super VL finishes tasks faster than competing models from Anthropic and OpenAI, including Opus 5, GPT 6.1 Sol, Luna and Astra, when post-trained on specific agentic tasks. These tasks include document intelligence, financial accounting, scientific experimentation, medical insurance claims and security attack resilience.
The company backs these claims with a list of major clients using Super VL. This list includes Skydance, EY, Ramp, Crowdstrike, Perplexity and Salesforce.
These are not the first open-weight models from Nvidia to compete with proprietary frontier labs. The GPU giant has a full family of Nemotron models covering multimodal artificial intelligence, reasoning, coding, math and enterprise automation. However, those fast-growing proprietary labs are now looking at other chips and system makers beyond Nvidia, such as AWS’s Trainium and Cerebras, as well as building their own hardware.
Nvidia is therefore leaning into the trend of enterprise companies running their own open models in their own data centres. Unlike closed frontier models, organisations can post-train Super VL on their own data, deploy it on their own infrastructure and own what they build, Briski said.
Local agents for consumers
With the announcement of the smaller, local PC models, Nvidia is also pushing forward the nascent trend of consumer AI assistants. While a new AI assistant appears daily from startups and big tech, privacy-weary consumers may not want to invite such agents deeply into their lives, giving them access to information about families, health and finances.
Nvidia hopes to create a new market for itself by selling PCs powered by its chips. These machines run models and agents locally. PC users can delegate everyday projects without sending those workloads to the cloud, Briski wrote.
Of the two Bolt models, the Nemotron 3.5 Bolt 74B-A9B 24GB contains 74 billion parameters. It is designed for local agentic tasks, balancing speed with reasoning to handle multistep reasoning, tool calling and web browsing. The smallest Bolt E4B is aimed at conversational agents.
Super VL and Bolt 74B-A9B are available now. Nvidia says the Bolt E4B will arrive later this month.
