Llama Guard 3 8B

An 8-billion-parameter content safety classifier from Meta that detects harmful content in text across multiple risk categories.

Strengths

Designed specifically for content moderation tasks, evaluating safety across categories like violence, illegal activity, harassment, and self-harm.

Best for

Screening user inputs or model outputs for harmful content in production applications that need safety filtering.

Limitations

A specialized safety classifier, not a general-purpose language model — it should not be used for tasks like text generation, question-answering, or reasoning; Llama Guard 4 12B (released April 2025) is a newer successor with expanded capabilities.