NVIDIA· 4 JunNemotron 3.5 Content Safety: Customizable Multimodal Safety for Global Enterprise AI
Single model unifies multimodal input, multilingual reach, custom policies, and auditable reasoning.
New model
NVIDIA released Nemotron 3.5 Content Safety, a single 4B-parameter model built on Google Gemma 3 4B IT that unifies multimodal input, multilingual coverage across 12 explicitly trained languages plus zero-shot generalization to ~140 languages, custom enterprise policy enforcement, and optional auditable reasoning traces.
The model accepts a user prompt, optional image, and optional assistant response in one context window and produces a safety verdict.
NVIDIA also released the multimodal, multilingual safety dataset used to train the model.
