NVIDIA releases Nemotron 3.5 Super VL
An open 120B model for multimodal agents, available on Hugging Face, ModelScope, and as an NVIDIA NIM.
Official source (opens in a new tab)
NVIDIA's 15 October 2026 post announces Nemotron 3.5 Super VL, an open 120-billion-parameter model for specialized always-on agents. NVIDIA says its accuracy on instruction following, tool use, multistep workflows, and visual understanding is comparable to NVIDIA's largest reasoning model at about one-fifth the parameter count, and that it is the fastest open agentic model in its class. The model reads web pages, PDFs, tables, and charts. It ships with open weights, open datasets, and training recipes, and it is available on Hugging Face, Open Router, the ModelScope community, and as an NVIDIA NIM.
NVIDIA says it runs on DGX Spark, DGX Station, RTX PRO and RTX systems, Jetson, and in the data center. The same post also covers Nemotron 3.5 Bolt 74B, which already has its own note, and says Bolt E4B 4GB is still scheduled for 23 October.
Who it's for: teams that want an open multimodal agent model they can post-train and run on their own NVIDIA hardware.
Fits: Best Agent Builder. It does not change the order of that list.