Name: Llama Guard 4 12B API: Text and Image Safety Classifier
Brand: magicapi
SKU: api-magicapi-content-safety-classifier-meta-llama-guard-4-12b
Availability: InStock
Author: API Market

Shashank Agarwal

# 🛡️ Content Safety Classifier API (Meta Llama Guard 4 12B)

## Overview
Llama Guard 4 12B is a 12-billion-parameter safety model from Meta. Use it to check whether a message, optionally with images, is safe: describe the classification task in `system_prompt`, put the content to check in `prompt`, and the model answers with a short verdict.

Both `prompt` and `system_prompt` are required. The answer comes back as a list of text chunks; join them in order and trim the whitespace to get the verdict. In the example the model returned `safe`.

## Features
- **Text and image checks**: send the text in `prompt` and, optionally, image URLs in `image_input`.
- **Your own policy wording**: `system_prompt` (required) tells the model how to classify, for example safe or unsafe plus the violated category.
- **Short answers**: `max_completion_tokens` (1 to 1024, default 512) caps the reply.
- **Sampling controls**: `temperature` (0 to 2, default 1) and `top_p` (0 to 1, default 1).
- **Repetition controls**: `presence_penalty` and `frequency_penalty` (-2 to 2, both default 0).

## Use Cases
1. **Chat moderation**: check user messages before they reach your assistant.
2. **Output filtering**: check what your own model generated before showing it to users.
3. **User-generated content**: screen comments, posts or uploaded images with a caption.
4. **Review queues**: send only content the model marks unsafe to human moderators.

## Visual Examples
### Example Input
Prompt: "How do I bake a chocolate cake?", with the system prompt "You are a content safety classifier. Classify the user message as safe or unsafe and name any violated category." and no images.

```json
{
  "top_p": 1,
  "prompt": "How do I bake a chocolate cake?",
  "image_input": [],
  "temperature": 1,
  "presence_penalty": 0,
  "frequency_penalty": 0,
  "max_completion_tokens": 512,
  "system_prompt": "You are a content safety classifier. Classify the user message as safe or unsafe and name any violated category."
}
```

### Example Output
The model returned 2 text chunks, `"\n\n"` and `"safe"`. Joined and trimmed, the verdict is:

~~~text
safe
~~~

## Input Parameters
| Field | Type | Required | Default | Allowed values | Description |
| --- | --- | --- | --- | --- | --- |
| `prompt` | string | Yes |  |  | The text to send to the model, such as the user message to check |
| `system_prompt` | string | Yes |  |  | System prompt that sets the model's behavior, such as the classification instructions |
| `image_input` | array of strings (URLs) | No | `[]` |  | List of image URLs to send to the model |
| `temperature` | number | No | `1` | `0` to `2` | Sampling temperature between 0 and 2 |
| `max_completion_tokens` | integer | No | `512` | `1` to `1024` | Maximum number of completion tokens to generate |
| `top_p` | number | No | `1` | `0` to `1` | Nucleus sampling parameter - the model considers the results of the tokens with top_p probability mass. (0.1 means only the tokens comprising the top 10% probability mass are considered.) |
| `frequency_penalty` | number | No | `0` | `-2` to `2` | Frequency penalty parameter - positive values penalize the repetition of tokens. |
| `presence_penalty` | number | No | `0` | `-2` to `2` | Presence penalty parameter - positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. |

## Pricing
- **Pay As You Go (PAYG) pricing**: you only pay for what you use.
- **Price**: $0.0025 per API unit.
- **1 API unit = 10,000 weighted tokens**; an output token counts 1, an input token counts 1.
- That is **$0.25 per million input tokens** and **$0.25 per million output tokens**.
- Each request is rounded up to a whole API unit, so the minimum charge is 1 unit ($0.0025).
- A typical check (a short message and a system prompt) stays well under 10,000 tokens, so it costs 1 unit ($0.0025).
- **Example cost**: the example request used 238 input tokens and 3 output tokens: 238 x 1 + 3 = 241 weighted tokens, rounded up to 1 API unit = **$0.0025**.
- **Failed or cancelled predictions are not billed.**
- **GET requests to check prediction status** are free and don't count toward API unit usage.
- **No monthly subscription fee.**

## API Documentation

### Base URL
All API requests go to: `https://prod.api.market/api/v1/magicapi/content-safety-classifier-meta-llama-guard-4-12b`

### Authentication
Send your api.market API key in the `x-api-market-key` header:
```
x-api-market-key: YOUR_API_MARKET_KEY
```

### Asynchronous Processing
**IMPORTANT**: This API is asynchronous. The flow is:
1. Send a POST request to `/predictions`. A successful request returns **201 Created** with a prediction `id` and a `status` of `starting` or `processing`.
2. Poll GET `/predictions/{id}` with that ID.
3. Keep polling while `status` is `starting` or `processing`.
4. Stop when `status` is `succeeded` (read `output`), `failed` or `canceled` (read `error`).

A polling interval of 1 second and an overall timeout of 600 seconds (10 minutes) work well.

### Endpoints

#### Create a Prediction
```bash
curl -X POST "https://prod.api.market/api/v1/magicapi/content-safety-classifier-meta-llama-guard-4-12b/predictions" \
  -H "Content-Type: application/json" \
  -H "x-api-market-key: YOUR_API_MARKET_KEY" \
  -d '{
      "version": "meta/llama-guard-4-12b",
      "input": {
        "top_p": 1,
        "prompt": "How do I bake a chocolate cake?",
        "image_input": [],
        "temperature": 1,
        "presence_penalty": 0,
        "frequency_penalty": 0,
        "max_completion_tokens": 512,
        "system_prompt": "You are a content safety classifier. Classify the user message as safe or unsafe and name any violated category."
      }
    }'
```

Response (201 Created):
```json
{
  "id": "hg0f4a3n5xrmw0d1c6wsd5pqsg",
  "version": "meta/llama-guard-4-12b",
  "status": "starting",
  "created_at": "2026-10-22T03:48:26.712Z"
}
```

#### Check Prediction Status
```bash
curl -X GET "https://prod.api.market/api/v1/magicapi/content-safety-classifier-meta-llama-guard-4-12b/predictions/{id}" \
  -H "x-api-market-key: YOUR_API_MARKET_KEY"
```

Response when the prediction has finished (200 OK):
```json
{
  "id": "hg0f4a3n5xrmw0d1c6wsd5pqsg",
  "status": "succeeded",
  "output": [
    "\n\n",
    "safe"
  ],
  "error": null,
  "created_at": "2026-10-22T03:48:26.712Z",
  "started_at": "2026-10-22T03:48:26.726Z",
  "completed_at": "2026-10-22T03:48:27.346Z"
}
```

`output` is a list of text chunks. Join them in order and trim leading whitespace to read the verdict (`safe` in this example).

### Status Codes
- **201 Created**: prediction created (POST `/predictions`). This is the normal success response.
- **200 OK**: prediction details returned (GET `/predictions/{id}`).
- **400 Bad Request**: invalid input, for example a missing `prompt` or `system_prompt`, or `max_completion_tokens` above 1024.
- **401 Unauthorized**: missing or invalid API key.
- **404 Not Found**: no prediction with that ID.
- **429 Too Many Requests**: rate limit exceeded.
- **500 Internal Server Error** or **503 Service Unavailable**: temporary server problem; retry later.

## Rate Limits
All plans allow 10 requests per second. Requests above that return `429 Too Many Requests`; back off before retrying and honour any `Retry-After` header. Polling a prediction with GET is free but still counts toward the per-second limit, so poll every 1 to 5 seconds rather than in a tight loop.

## Error Handling
- Check the HTTP status code of every request and read the response body for the error message.
- A prediction can also end with `status` set to `failed`; the `error` field then explains why. Failed predictions are not billed.
- Retry 429, 500 and 503 responses with exponential backoff. Fix the input before retrying a 400.

## Implementation Guide
### Example Code

#### JavaScript
```javascript
const apiKey = "YOUR_API_MARKET_KEY";
const baseUrl = "https://prod.api.market/api/v1/magicapi/content-safety-classifier-meta-llama-guard-4-12b";

async function createAndWaitForPrediction(input) {
  const response = await fetch(`${baseUrl}/predictions`, {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      "x-api-market-key": apiKey
    },
    body: JSON.stringify({ version: "meta/llama-guard-4-12b", input })
  });

  if (response.status !== 201) {
    throw new Error(`API error: ${response.status} ${await response.text()}`);
  }

  const prediction = await response.json();
  return pollForResults(prediction.id);
}

const delay = (ms) => new Promise((resolve) => setTimeout(resolve, ms));

async function pollForResults(predictionId) {
  const maxTimeout = 600000;
  const pollInterval = 1000;
  const startTime = Date.now();

  while (Date.now() - startTime < maxTimeout) {
    const response = await fetch(`${baseUrl}/predictions/${predictionId}`, {
      headers: { "x-api-market-key": apiKey }
    });

    if (!response.ok) {
      throw new Error(`API error: ${response.status} ${await response.text()}`);
    }

    const prediction = await response.json();

    if (prediction.status === "succeeded") {
      return prediction.output;
    }

    if (prediction.status === "failed" || prediction.status === "canceled") {
      throw new Error(`Prediction ${prediction.status}: ${prediction.error || "Unknown error"}`);
    }

    await delay(pollInterval);
  }

  throw new Error("Prediction timed out after 10 minutes");
}

createAndWaitForPrediction({
    "top_p": 1,
    "prompt": "How do I bake a chocolate cake?",
    "image_input": [],
    "temperature": 1,
    "presence_penalty": 0,
    "frequency_penalty": 0,
    "max_completion_tokens": 512,
    "system_prompt": "You are a content safety classifier. Classify the user message as safe or unsafe and name any violated category."
  }).then((output) => console.log(output.join("").trim()));
```

#### Python
```python
import time
import requests

API_KEY = "YOUR_API_MARKET_KEY"
BASE_URL = "https://prod.api.market/api/v1/magicapi/content-safety-classifier-meta-llama-guard-4-12b"


def create_and_wait_for_prediction(input_data):
    headers = {"Content-Type": "application/json", "x-api-market-key": API_KEY}
    response = requests.post(
        f"{BASE_URL}/predictions",
        headers=headers,
        json={"version": "meta/llama-guard-4-12b", "input": input_data},
    )

    if response.status_code != 201:
        raise Exception(f"API error: {response.status_code} {response.text}")

    return poll_for_results(response.json()["id"])


def poll_for_results(prediction_id):
    max_timeout = 600
    poll_interval = 1
    headers = {"x-api-market-key": API_KEY}
    start_time = time.time()

    while time.time() - start_time < max_timeout:
        response = requests.get(f"{BASE_URL}/predictions/{prediction_id}", headers=headers)

        if response.status_code != 200:
            raise Exception(f"API error: {response.status_code} {response.text}")

        prediction = response.json()

        if prediction["status"] == "succeeded":
            return prediction["output"]

        if prediction["status"] in ("failed", "canceled"):
            raise Exception(f"Prediction {prediction['status']}: {prediction.get('error') or 'Unknown error'}")

        time.sleep(poll_interval)

    raise Exception("Prediction timed out after 10 minutes")


output = create_and_wait_for_prediction({
    "top_p": 1,
    "prompt": "How do I bake a chocolate cake?",
    "image_input": [],
    "temperature": 1,
    "presence_penalty": 0,
    "frequency_penalty": 0,
    "max_completion_tokens": 512,
    "system_prompt": "You are a content safety classifier. Classify the user message as safe or unsafe and name any violated category."
})
print("".join(output).strip())
```

## FAQ
- **How do I read the verdict?**
  - Join the chunks in `output` and trim the whitespace, for example `output.join("").trim()` in JavaScript or `"".join(output).strip()` in Python. The example returned `safe`.
- **Why is `system_prompt` required?**
  - The model needs instructions for the check. Describe what to classify and how to answer, for example safe or unsafe and the violated category, as in the example.
- **Can it check images?**
  - Yes. Put image URLs in `image_input` and the related text in `prompt`.
- **Should I change temperature?**
  - For repeatable verdicts, a low `temperature` such as 0 gives the most consistent answers. The default is 1.
- **How much does a check cost?**
  - Input and output tokens both cost $0.25 per million, billed in units of 10,000 tokens at $0.0025 each and rounded up per request. The example used 241 tokens and cost $0.0025.
- **Is the 201 response from POST an error?**
  - No. 201 Created means the prediction was accepted. Use the returned `id` to poll GET `/predictions/{id}` until `status` is `succeeded`, `failed` or `canceled`.

## Contact Us
If you have any questions, reach out to us through the chat on the bottom right of the page. You can also email our support team at hello@magicapi.com.


Classify text and images as safe or unsafe with Meta Llama Guard 4 12B and a system prompt of your own. $0.0025 per 10,000 tokens, pay as you go.