Llama Guard 4 12B
🧠 AI Modelmeta-llama
Multimodal content safety classifier based on Llama 4 Scout, fine-tuned for safe LLM interactions.
Llama Guard 4 12B builds on Meta's Llama 4 Scout architecture, adding safety classification capabilities for both text and image inputs. With a 163,840 token context window, it can analyze large content blocks. It serves as a guardrail for LLM outputs, identifying harmful categories. The model is available via OpenRouter with pay-as-you-go pricing ($0.18/M tokens for both input and output), and supports features like frequency penalty, logit bias, and response format.
💡Highlights
- ├─12B parameters multimodal model
- ├─163,840 token context
- └─Image and text safety classification
🎯For
- ├─AI safety researchers
- ├─LLM developers
- └─content moderation teams