Nvidia on Thursday announced the launch of a new family of open-weight models, the Nemotron 3.5 series, aimed at powering AI agents across enterprise data centers and personal computers.
At the high end, the Nemotron 3.5 Super VL is a 120-billion-parameter model designed for always-on enterprise agents. According to Nvidia, it delivers similar accuracy to its larger Nemotron 3 Ultra model in instruction-following and tool-use tasks, while using fewer parameters.
On the other end of the spectrum, the company released two lightweight Bolt models intended to run locally on the new generation of AI PCs equipped with Nvidia's RTX Spark chip or workstations running DGX Spark. These PCs, marketed with partners including Microsoft and Dell, are aimed at consumers, programmers and enterprise knowledge workers.
The Bolt models come in two sizes: Nemotron 3.5 Bolt 74B-A9B 24GB, a 74-billion-parameter model designed for local agentic tasks such as multistep reasoning, tool calling and web browsing; and a smaller Bolt E4B aimed at conversational agents. Super VL and the larger Bolt model are available now, with Bolt E4B coming later this month, Nvidia said.
"Any organization that wants to run agents on Nvidia infrastructure now has an open, optimized model for their tier without having to choose between capability and deployability," Kari Briski, Nvidia's vice president of Open Models and Software, wrote in a blog post announcing the models.
Nvidia claims that Super VL completes tasks faster than competing models from Anthropic and OpenAI when post-trained on specific agentic tasks such as document intelligence, financial accounting, scientific experimentation, medical insurance claims and security attack resilience. Such comparative claims are standard when major tech companies introduce new models. Nvidia has backed its claims with a list of companies it says are using Super VL, including Skydance, EY, Ramp, Crowdstrike, Perplexity and Salesforce.
The GPU giant already has a family of Nemotron models covering multimodal AI, reasoning, coding, math and enterprise automation. However, the company faces increasing competition from alternative chipmakers such as AWS's Trainium and Cerebras, as well as proprietary chips built by AI labs themselves. By pushing open-weight models, Nvidia aims to serve enterprises that want to run their own models on their own infrastructure.
"Unlike closed frontier models, organizations can post-train Super VL on their own data, deploy it on their own infrastructure and own what they build," Briski wrote.
With the smaller Bolt models, Nvidia is also trying to advance the trend of consumer AI assistants that can run locally on PCs. Briski noted that users "can delegate everyday projects ... without sending those workloads to the cloud," addressing privacy concerns around cloud-based AI agents.