Llama Guard
Llama Guard classifies content in large language model inputs and responses as safe or unsafe.
What Llama Guard is
Llama Guard classifies content in large language model inputs and responses as safe or unsafe. The Llama Guard 3-8B model is a pretrained Llama 3.1 8B model fine-tuned for content safety classification. It covers 14 hazard categories, including the 13 MLCommons categories and Code Interpreter Abuse, and supports English, French, German, Hindi, Italian, Portuguese, Spanish, and Thai. The model is optimized to assess search and code interpreter tool calls. Llama Recipes documentation describes configuration and customization, while a Llama Cookbook example uses Hugging Face Transformers to check user prompts and model outputs; that demo requires access to the model weights on Hugging Face. The model is free and open source for self-hosted deployment. An int8 quantized version reduces checkpoint size by about 40%, with very small impact on performance according to the model card. Meta recommends deploying it alongside Llama 3.1 and cautions that it may increase refusals of benign prompts. The model card also notes limitations related to pretraining data and possible vulnerability to adversarial or prompt injection attacks. Some categories, including defamation, intellectual property, and elections, may require current factual knowledge for accurate evaluation.
Good
- Covers 14 hazard categories
- Supports eight listed languages
- Optimized for search and code interpreter tool calls
- Can be customized using Llama Recipes documentation
Watch out
- Requires self-hosted deployment
- May refuse some benign prompts
- May be vulnerable to adversarial or prompt injection attacks
- Some categories need current factual knowledge
Who it suits
Llama Guard suits teams seeking a self-hosted content safety classifier for LLM inputs and responses. Its language coverage and tool-call focus may fit applications that need those specific checks.
Verdict
Llama Guard offers a self-hosted classifier with defined category and language coverage. Account for its stated refusal and evaluation limits when deciding whether it fits a safety workflow.
Plans and prices
From Llama Guard's own pricing page, checked 4 Oct 2026.
| Plan | Price | What it gives you |
|---|---|---|
| Llama Guard 3-8B | Free | 8B parameters · model weights required |
The facts
Each one read from Llama Guard's own pages, with the date it was checked.
What it does
- Customization
- Llama Recipes documentation explains how to configure and customize Llama Guard 3.checked 4 Oct 2026
- Evaluation caveat
- The model card notes that some categories, including defamation, intellectual property, and elections, may require factual, up-to-date knowledge for accurate evaluation.checked 4 Oct 2026
- Hazard categories
- Llama Guard 3-8B covers 14 categories, including the 13 MLCommons hazard categories and Code Interpreter Abuse.checked 4 Oct 2026
- Integration example
- Meta’s Llama Cookbook demonstrates using Llama Guard as an inference safety checker for both user prompts and model outputs with Hugging Face Transformers.checked 4 Oct 2026
- Languages
- It supports content safety classification in English, French, German, Hindi, Italian, Portuguese, Spanish, and Thai.checked 4 Oct 2026
- Limitations
- The model card says performance can be limited by pretraining data and that the model may be vulnerable to adversarial or prompt injection attacks.checked 4 Oct 2026
- Model
- Llama Guard 3-8B is a Llama 3.1 8B pretrained model fine-tuned for content safety classification.checked 4 Oct 2026
- Purpose
- Llama Guard classifies content in LLM inputs and responses as safe or unsafe.checked 4 Oct 2026
- Quantization
- An int8 quantized version reduces checkpoint size by about 40% with very small impact on model performance, according to the model card.checked 4 Oct 2026
- Recommended use
- Meta recommends deploying Llama Guard 3 alongside Llama 3.1, and says it may increase refusals of benign prompts.checked 4 Oct 2026
- Tool use
- It was optimized for safety and security classification of search and code interpreter tool calls.checked 4 Oct 2026
- Weights access
- The Cookbook demo lists access to Llama Guard model weights on Hugging Face as a requirement.checked 4 Oct 2026
Alternatives to Llama Guard
- 1
Google Cloud Vision API
Free planWebAPILinuxmacOS7.9$0.60 usage-based - 2
Sightengine
Free planWebAPISelf-hosted7.5From $29 a month - 3
WebPurify
Free planAndroidiOSWebAPI7.4From $5 a month - 4
Amazon Rekognition Content Moderation
Free planWebAPI7.3From $0.10 a month - 5
Tisane API
Free planWebAPILinuxmacOS7.1From $99 a month - 6
Alibaba Cloud Content Moderation
iOSWebAPI7.0$0.09 usage-based