Nvidia releases Nemotron 3.5 Super VL, an open 120-billion-parameter vision model for agents, and a Bolt model that runs locally in 24GB

Nvidia released Nemotron 3.5 Super VL on 15 October, describing it on its blog as “an open 120-billion-parameter model”. The model reads documents, charts and other visual material for AI agents. Nvidia says it is the fastest open agentic model in its class, five times faster than the next best; that is Nvidia's own claim and has not been checked independently.

The model ships with open weights, open datasets and training recipes. The blog names no licence and does not give the active parameter count in its text. It says the model is available now on Hugging Face, OpenRouter, ModelScope and as an Nvidia NIM service.

Nvidia also released Nemotron 3.5 Bolt 74B-A9B, which uses native 2-bit quantization to fit in 24GB of memory so it can run on a local machine. A smaller model, listed as Bolt E4B 4GB, is due on 23 October. Nvidia names Atlassian, CrowdStrike, Perplexity, Salesforce and Synopsys among the companies integrating the models.

The release came the day before Meta published the open weights of Muse Spark Open, which the Switchboard ran on 16 October.

• Confirmed Nvidia released Nemotron 3.5 Super VL, an open 120-billion-parameter model, available on Hugging Face, OpenRouter, ModelScope and NIM. Nvidia blog
• Claimed Nvidia says the model is 5x faster than the next best open model in its class. Nvidia blog
• Confirmed Nemotron 3.5 Bolt 74B-A9B fits in 24GB; Bolt E4B is due on 23 October. Nvidia blog