Nemotron 3.5 Content Safety is NVIDIA’s compact multimodal safety and guardrail model for moderating prompts, optional images, and optional assistant responses in a single request. It is designed for content-safety classification across text and image inputs, with support for custom policy enforcement and reasoning-based moderation.
The model is presented as a multilingual safety moderator built on Google Gemma 3 4B IT, and NVIDIA describes it as fine-tuned for multimodal and multilingual content-safety tasks.
| Developer | NVIDIA |
|---|---|
| Model ID | nvidia/nemotron-3.5-content-safety |
| Release Date | June 4, 2026 |
| Context Window | 131,072 tokens (~131K) |
| Max Output Tokens | 117,964 tokens |
| Modalities | Image, Text → Text |
| Capabilities | streaming, vision |
Multimodal moderation
The model accepts a prompt, an optional image, and an optional response, then returns safety labels for the user input and response if present.
Custom policy support
It can take a user-defined safety policy and, in custom-policy mode, produce a concise reasoning trace before the final classification.
Multilingual coverage
The model supports 12 languages according to NVIDIA and third-party model listings.
Safety taxonomy
In standard mode, it classifies content using NVIDIA’s safety taxonomy and can return violated safety categories.
Base model and size
NVIDIA describes it as a small language model built on Google Gemma-3-4B-it / Gemma 3 4B IT, with a 4B-parameter multimodal architecture.
Prompt and response moderation
Use it to score user prompts and assistant replies as safe or unsafe before content is shown or acted on.
Image-aware safety filtering
Use it to moderate text-plus-image interactions where a single image may affect the safety decision.
Enterprise policy enforcement
Use it when you need developer-defined moderation rules rather than only a fixed safety taxonomy.
Audit-friendly moderation workflows
Use its reasoning mode when you need a short explanation trace to support review or policy tuning.
Nemotron 3 Content Safety Multimodal
NVIDIA’s NGC Catalog says Nemotron 3.5 Content Safety preserves the multimodal moderation behavior of Nemotron 3 Content Safety while adding custom-policy adaptation.
Nemotron Content Safety Reasoning 4B
NVIDIA’s NGC Catalog says Nemotron 3.5 Content Safety unifies the multimodal model with the custom-policy capabilities of the Nemotron Content Safety Reasoning 4B model.
Sources: https://huggingface.co/blog/nvidia/nemotron-3-5-content-safety · https://build.nvidia.com/nvidia/nemotron-3.5-content-safety · https://blogs.vultr.com/nemotron-3-5-content-safety
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
Used as a content-safety moderator for inputs to and responses from LLMs and VLMs, including checking a prompt, an optional image, and an optional response for safe versus unsafe content.
Suited for safety and moderation workflows, where it detects unsafe and toxic content in multilingual, multimodal applications.
Used for multimodal content-safety moderation in enterprise AI guardrails, including custom-policy enforcement and reasoning-trace classification.
Applicable to prompt and response moderation pipelines, including real-time moderation of user inputs and model outputs.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "nvidia/nemotron-3.5-content-safety", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "nvidia/nemotron-3.5-content-safety", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"nvidia/nemotron-3.5-content-safety","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Nemotron 3.5 Content Safety This page | Vision and document understanding | |||
| Coding + agents | ||||
| Coding + agents |
The model has a 131,072-token context window and can return up to 117,964 output tokens in a single response.
$0.27508 per 1M input tokens and $0.27508 per 1M output tokens on AI/ML API. Input and output are priced the same.
It accepts image and text input and returns text. That lets a single request carry a prompt, an optional image, and an optional response to be judged together.
Yes. Streaming and vision are supported.
June 4, 2026.
Use nvidia/nemotron-3.5-content-safety as the model id.
Send a request to the chat completions endpoint with the model id nvidia/nemotron-3.5-content-safety. The request shape is the same as for any other chat model on AI/ML API.
