Nemotron 3.5 Content Safety: Customizable Multimodal Safety for Global Enterprise AI

NVIDIA released Nemotron 3.5 Content Safety, a 4B-parameter model that unifies multimodal input processing, multilingual support across 12 languages, and customizable enterprise policy enforcement in a single inference call. The model evaluates user prompts, images, and assistant responses together to catch policy violations emerging from text-image interactions, while offering optional reasoning traces for audit purposes. Key improvements over its predecessor include custom policy specification at inference time, efficient reasoning outputs, and strong performance across multilingual and multimodal safety benchmarks.