Llama Guard 4 12B
Open multimodal safety classifier for prompts and responses.
Llama Guard 4 is derived from Llama 4 Scout and fine-tuned for content-safety classification. It classifies both prompts (input) and responses (output) — including images — against a hazard taxonomy, returning safe/unsafe plus the violated categories.
- Provider
- Meta
- Type
- Safety classifier
- Released
- Apr 29, 2025
- Context window
- 164K tokens
- Max output
- 16K tokens
- Price
- $0.18 input / $0.18 output per 1M tokens
- Input
- text, image
- Output
- text
- Open weights
- Yes
- License
- Llama 4 Community License
Best for
- Enterprise
- Extraction & classification
- Vision
Strengths
- Open weights
- Text + image moderation
- Standard hazard taxonomy
More from Meta
- Llama 4 Maverick — Meta. Meta’s 128-expert open multimodal MoE model.
- Llama 4 Scout — Meta. Efficient open model with an industry-leading 10M-token native context.
- Muse Spark 1.3 — Meta. Meta’s multimodal reasoning model for long-running agents and coding.
- Muse Image — Meta. Meta’s low-cost image model.
- V-JEPA 2 — Meta. Meta’s open JEPA world model for video understanding, prediction and robot planning.
Tools that use it
- LlamaFirewall — Meta. Open-source guardrail system for AI agents.