Llama Guard 4 12B
Neural Network
Llama Guard 4 12B is Meta's multimodal content classification model, a Llama Guard 4 API for checking LLM requests and responses.
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 163 840 tokens
- Max answer length
- 16 384 tokens
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works Llama Guard 4 12B?
Llama Guard 4 is a 12B parameter safety classifier from Meta, built on Llama 4 Scout. The model does not hold a dialogue but issues a verdict: it labels material by risk categories and acts as a filter on both sides of the conversation, checking both the user request and the main model's response. Multimodality allows it to analyze not only text but also images, covering mixed content moderation in a single loop. In BotHub, Llama Guard 4 connects via a unified OpenAI-compatible API alongside other models, so the filtering layer and generative layer live in one integration, and switching models does not require rewriting code. Access is open from Russia without a VPN or foreign card; pay in rubles for tokens used, and correspondence is encrypted. For teams, there is a contract, invoice, and admin panel with limits. It is useful for protecting support chatbots, filtering risky prompts in products with user input, labeling datasets, auditing dialogue logs, and checking uploaded images.Frequently asked questions about Llama Guard 4 12B
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
llama-guard-4-12b is a Meta model for content safety classification: it evaluates requests and responses and marks them as safe or policy-violating. It is suitable for moderating user input, filtering output from other models, and checking documents and images before publication.
The model outputs up to 16,384 tokens per response, and the context window is 163,840 tokens, so you can submit a long dialogue or an entire document for verification. In practice, the response is short: the model returns a compact verdict rather than a detailed text.
In our data, a separate reasoning mode for llama-guard-4-12b is not noted. It is a classifier: it analyzes the submitted fragment and immediately issues a verdict indicating the violation category, without long chains of thought. If you need step-by-step reasoning, you can switch to another model in BotHub in the same window.
In our data, function calling and structured JSON output for llama-guard-4-12b are not noted. However, the model's response is concise and easily parsed on your service's side. Connection is via the unified OpenAI-compatible BotHub API, so integrating the check into an existing pipeline is simple.
Images and documents are supported as input along with text, so you can send images and files for verification in a single request. Audio and video are not declared in our data. Image or sound generation is not provided here: only text is output.
We do not have data on the languages that llama-guard-4-12b works with, and we will not speculate. Test the model with your own examples: in BotHub, access is open from Russia without a VPN or foreign card, and payment is based on actual tokens used.