Nvidia releases Nemotron models for enterprise and local agents

NVIDIA says Super VL is a 120-billion-parameter open model designed for specialized, always-on agents.[1] The release also includes Bolt 74B-A9B, which uses 24GB of memory, and Bolt E4B, which uses 4GB.[1] NVIDIA positions the three models across enterprise servers, edge devices and local hardware, rather than as a single model for one deployment setting.[1] For teams assessing options, the announced lineup separates a large model for always-on agent work from two models aimed at more constrained local or edge environments.[1]

NVIDIA says Super VL delivers accuracy on multimodal agentic tasks that include instruction following, tool use, multistep business workflows and visual understanding.[1] The company says it is comparable with its largest reasoning model at roughly one-fifth the parameter count, and five times faster than the next-best model in its class.[1] Super VL was trained on about 26 trillion tokens and post-trained with roughly 3.8 million samples, according to NVIDIA.[1]

NVIDIA says Bolt 74B-A9B is designed for everyday agentic tasks running locally on RTX or DGX Spark systems.[1] Its native 2-bit quantization is intended to fit the model in 24GB of memory while allowing it to run alongside other applications.[1] NVIDIA says local projects could include web research, writing and code debugging without sending workloads to the cloud.[1] For organizations choosing Super VL, NVIDIA says the model comes with open weights, datasets and training recipes; teams can post-train it on their own data and deploy it on their own infrastructure.[1]

NVIDIA says Super VL and Bolt 74B-A9B are available immediately, while Bolt E4B is due October 23.[1] Super VL is offered through Hugging Face, Open Router, ModelScope community and NVIDIA NIM, and NVIDIA says it runs on systems including DGX Spark, DGX Station, RTX PRO and RTX, Jetson, and data-center and cloud environments.[1] NVIDIA says Scale AI post-trained Super VL for agents for Skydance vendor-credit finance and EY M&A due diligence; Ramp and CrowdStrike are customizing it for financial document intelligence and enterprise-security blue teaming, respectively.[1]