Nemotron 3.5 Content Safety

NVIDIA's NIM product page for nemotron-3.5-content-safety offers a free, rate-limited trial endpoint for the multilingual, multimodal safety model. The page tags the model under ai safety, nemo guardrails, llm safety, multilingual content safety, and safety and moderation, and is accelerated by DGX Cloud. The trial service is governed by the NVIDIA API Trial Terms of Service, while model use is governed by the NVIDIA Nemotron Open Model License plus the Gemma Terms of Use and Gemma Prohibited Use Policy. Developers access the model through build.nvidia.com with a free API key for development, with traffic subject to throttling and rate limits shared across users. The page provides API reference and deployment links alongside a model card and example inputs, but detailed architecture and benchmark specifics live on the companion model card page. This listing establishes the subject model's official NIM endpoint and licensing posture for product integration.

NVIDIA's NIM API reference documents the Nemotron 3.5 Content Safety model as a small language model fine-tuned by NVIDIA on top of Google's Gemma-3-4B-it base, with a LoRA adapter merged back into the base weights. It accepts a prompt, an optional image, and an optional response, returning safety labels and a list of violated categories when content is flagged as unsafe. The model targets multimodal content moderation for both LLM and VLM inputs and outputs, using the same taxonomy as the prior Nemotron 8B content-safety model. The page confirms release dates of 2026-03-16 on Hugging Face and NGC and 2026-06-02 on build.nvidia.com, with global deployment geography. Use is governed by the OpenMDW-1.1 license, Gemma Terms of Use, and the Gemma Prohibited Use Policy. The model is ready for commercial use and is positioned as an extension of the popular Nemotron 8B content-safety model, adding multimodal capability to text-only moderation.