NVIDIA releases open Nemotron 3.5 Super VL and Bolt models for agentic work

NVIDIA

NVIDIA has released Nemotron 3.5 Super VL, an open 120-billion-parameter multimodal model it calls the fastest open agentic model in its class, alongside Bolt models built for local use on RTX and DGX Spark hardware. The Super VL model is available on Hugging Face, OpenRouter, ModelScope and as an NVIDIA NIM.

• NVIDIA reports about 5x the throughput of the next-best model in its class, with accuracy close to its largest reasoning model at roughly one-fifth the parameters.
• Super VL was pretrained on about 26 trillion tokens, then post-trained on roughly 3.8 million samples and reinforcement learning.
• Bolt 74B-A9B fits in 24GB of memory through native 2-bit quantization; NVIDIA ranks it first among open models in its class on PinchBench.
• The 4GB Bolt E4B is due on Oct. 23; the article says the models ship with open weights, datasets and training recipes but names no license.