llama-guard-4-12b

llama-guard-4-12b: Moderations

curl --request POST \
  --url https://api.zerogpu.ai/v1/moderations \
  --header 'Content-Type: application/json' \
  --header 'x-api-key: <api-key>' \
  --data '
{
  "model": "llama-guard-4-12b",
  "input": "I am so angry at this person that I want to hurt them. They are worthless and should be scared of what I might do next."
}
'

import requests

url = "https://api.zerogpu.ai/v1/moderations"

payload = {
    "model": "llama-guard-4-12b",
    "input": "I am so angry at this person that I want to hurt them. They are worthless and should be scared of what I might do next."
}
headers = {
    "x-api-key": "<api-key>",
    "Content-Type": "application/json"
}

response = requests.post(url, json=payload, headers=headers)

print(response.text)

const options = {
  method: 'POST',
  headers: {'x-api-key': '<api-key>', 'Content-Type': 'application/json'},
  body: JSON.stringify({
    model: 'llama-guard-4-12b',
    input: 'I am so angry at this person that I want to hurt them. They are worthless and should be scared of what I might do next.'
  })
};

fetch('https://api.zerogpu.ai/v1/moderations', options)
  .then(res => res.json())
  .then(res => console.log(res))
  .catch(err => console.error(err));

package main

import (
	"fmt"
	"strings"
	"net/http"
	"io"
)

func main() {

	url := "https://api.zerogpu.ai/v1/moderations"

	payload := strings.NewReader("{\n  \"model\": \"llama-guard-4-12b\",\n  \"input\": \"I am so angry at this person that I want to hurt them. They are worthless and should be scared of what I might do next.\"\n}")

	req, _ := http.NewRequest("POST", url, payload)

	req.Header.Add("x-api-key", "<api-key>")
	req.Header.Add("Content-Type", "application/json")

	res, _ := http.DefaultClient.Do(req)

	defer res.Body.Close()
	body, _ := io.ReadAll(res.Body)

	fmt.Println(string(body))

}

require 'uri'
require 'net/http'

url = URI("https://api.zerogpu.ai/v1/moderations")

http = Net::HTTP.new(url.host, url.port)
http.use_ssl = true

request = Net::HTTP::Post.new(url)
request["x-api-key"] = '<api-key>'
request["Content-Type"] = 'application/json'
request.body = "{\n  \"model\": \"llama-guard-4-12b\",\n  \"input\": \"I am so angry at this person that I want to hurt them. They are worthless and should be scared of what I might do next.\"\n}"

response = http.request(request)
puts response.read_body

{
  "id": "modr-0a1b2c3d4e5f60718293a4b5c6d7e8f90",
  "model": "llama-guard-4-12b",
  "results": [
    {
      "flagged": true,
      "categories": {
        "harassment": true,
        "harassment/threatening": true,
        "hate": false,
        "hate/threatening": false,
        "illicit": false,
        "illicit/violent": false,
        "self-harm": false,
        "self-harm/intent": false,
        "self-harm/instructions": false,
        "sexual": false,
        "sexual/minors": false,
        "violence": true,
        "violence/graphic": false
      },
      "category_scores": {
        "harassment": 0.9412,
        "harassment/threatening": 0.9016,
        "hate": 0.0231,
        "hate/threatening": 0.0104,
        "illicit": 0.0057,
        "illicit/violent": 0.0039,
        "self-harm": 0.0021,
        "self-harm/intent": 0.0012,
        "self-harm/instructions": 0.0008,
        "sexual": 0.0006,
        "sexual/minors": 0.0002,
        "violence": 0.8774,
        "violence/graphic": 0.0311
      },
      "category_applied_input_types": {
        "harassment": [
          "text"
        ],
        "harassment/threatening": [
          "text"
        ],
        "hate": [
          "text"
        ],
        "hate/threatening": [
          "text"
        ],
        "illicit": [
          "text"
        ],
        "illicit/violent": [
          "text"
        ],
        "self-harm": [
          "text"
        ],
        "self-harm/intent": [
          "text"
        ],
        "self-harm/instructions": [
          "text"
        ],
        "sexual": [
          "text"
        ],
        "sexual/minors": [
          "text"
        ],
        "violence": [
          "text"
        ],
        "violence/graphic": [
          "text"
        ]
      }
    }
  ]
}

{}

POST

moderations

llama-guard-4-12b: Moderations

curl --request POST \
  --url https://api.zerogpu.ai/v1/moderations \
  --header 'Content-Type: application/json' \
  --header 'x-api-key: <api-key>' \
  --data '
{
  "model": "llama-guard-4-12b",
  "input": "I am so angry at this person that I want to hurt them. They are worthless and should be scared of what I might do next."
}
'

import requests

url = "https://api.zerogpu.ai/v1/moderations"

payload = {
    "model": "llama-guard-4-12b",
    "input": "I am so angry at this person that I want to hurt them. They are worthless and should be scared of what I might do next."
}
headers = {
    "x-api-key": "<api-key>",
    "Content-Type": "application/json"
}

response = requests.post(url, json=payload, headers=headers)

print(response.text)

const options = {
  method: 'POST',
  headers: {'x-api-key': '<api-key>', 'Content-Type': 'application/json'},
  body: JSON.stringify({
    model: 'llama-guard-4-12b',
    input: 'I am so angry at this person that I want to hurt them. They are worthless and should be scared of what I might do next.'
  })
};

fetch('https://api.zerogpu.ai/v1/moderations', options)
  .then(res => res.json())
  .then(res => console.log(res))
  .catch(err => console.error(err));

package main

import (
	"fmt"
	"strings"
	"net/http"
	"io"
)

func main() {

	url := "https://api.zerogpu.ai/v1/moderations"

	payload := strings.NewReader("{\n  \"model\": \"llama-guard-4-12b\",\n  \"input\": \"I am so angry at this person that I want to hurt them. They are worthless and should be scared of what I might do next.\"\n}")

	req, _ := http.NewRequest("POST", url, payload)

	req.Header.Add("x-api-key", "<api-key>")
	req.Header.Add("Content-Type", "application/json")

	res, _ := http.DefaultClient.Do(req)

	defer res.Body.Close()
	body, _ := io.ReadAll(res.Body)

	fmt.Println(string(body))

}

require 'uri'
require 'net/http'

url = URI("https://api.zerogpu.ai/v1/moderations")

http = Net::HTTP.new(url.host, url.port)
http.use_ssl = true

request = Net::HTTP::Post.new(url)
request["x-api-key"] = '<api-key>'
request["Content-Type"] = 'application/json'
request.body = "{\n  \"model\": \"llama-guard-4-12b\",\n  \"input\": \"I am so angry at this person that I want to hurt them. They are worthless and should be scared of what I might do next.\"\n}"

response = http.request(request)
puts response.read_body

{
  "id": "modr-0a1b2c3d4e5f60718293a4b5c6d7e8f90",
  "model": "llama-guard-4-12b",
  "results": [
    {
      "flagged": true,
      "categories": {
        "harassment": true,
        "harassment/threatening": true,
        "hate": false,
        "hate/threatening": false,
        "illicit": false,
        "illicit/violent": false,
        "self-harm": false,
        "self-harm/intent": false,
        "self-harm/instructions": false,
        "sexual": false,
        "sexual/minors": false,
        "violence": true,
        "violence/graphic": false
      },
      "category_scores": {
        "harassment": 0.9412,
        "harassment/threatening": 0.9016,
        "hate": 0.0231,
        "hate/threatening": 0.0104,
        "illicit": 0.0057,
        "illicit/violent": 0.0039,
        "self-harm": 0.0021,
        "self-harm/intent": 0.0012,
        "self-harm/instructions": 0.0008,
        "sexual": 0.0006,
        "sexual/minors": 0.0002,
        "violence": 0.8774,
        "violence/graphic": 0.0311
      },
      "category_applied_input_types": {
        "harassment": [
          "text"
        ],
        "harassment/threatening": [
          "text"
        ],
        "hate": [
          "text"
        ],
        "hate/threatening": [
          "text"
        ],
        "illicit": [
          "text"
        ],
        "illicit/violent": [
          "text"
        ],
        "self-harm": [
          "text"
        ],
        "self-harm/intent": [
          "text"
        ],
        "self-harm/instructions": [
          "text"
        ],
        "sexual": [
          "text"
        ],
        "sexual/minors": [
          "text"
        ],
        "violence": [
          "text"
        ],
        "violence/graphic": [
          "text"
        ]
      }
    }
  ]
}

{}

Meta’s Llama Guard 4 12B is a multimodal safety classification model for moderating text, images, and mixed text-image inputs. It evaluates both incoming prompts and generated responses, returning a safe or unsafe classification along with the policy categories a violation falls under. It handles multilingual text and multiple images in one call, which makes it a fit for chat moderation, prompt and response filtering, policy enforcement, and agent guardrails. Built on a dense 12B architecture derived from Llama 4 Scout, it is designed as a dedicated safety layer in front of a production model rather than as a general-purpose generator.

References: Model docs • Terms • Privacy

Authorizations

x-api-key

string

header

required

Headers

x-project-id

string

Optional project identifier. Scopes the request to a specific project when provided.

Body

application/json

model

string

default:llama-guard-4-12b

required

Model identifier (fixed for this playground).

Allowed value: "llama-guard-4-12b"

Example:

"llama-guard-4-12b"

input

required

Text to classify. Accepts a single string, an array of strings (one result per element), or an array of content parts ({ "type": "text", "text": "..." }) that form a single multi-modal input.

Required string length: 1 - 131072

Response

Success

OpenAI-compatible moderations envelope. One results entry per input.

string

Unique identifier for the moderation request.

Example:

"modr-0a1b2c3d4e5f60718293a4b5c6d7e8f90"

model

string

The model used for classification.

Example:

"llama-guard-4-12b"

results

object[]

Moderation verdicts, one per input (a string input yields a single-element array).

Hide child attributes

results.flagged

boolean

True when the model flagged the input in one or more categories.

Example:

true

results.categories

object

Per-category boolean verdicts. All 13 categories are always present, in OpenAI's order.

Hide child attributes

results.categories.harassment

boolean

Content that expresses, incites, or promotes harassing language towards any target.

Example:

true

results.categories.harassment/threatening

boolean

Harassment that also includes violence or serious harm towards any target.

Example:

true

results.categories.hate

boolean

Content that expresses, incites, or promotes hate based on a protected attribute.

Example:

false

results.categories.hate/threatening

boolean

Hateful content that also includes violence or serious harm towards a protected group.

Example:

false

results.categories.illicit

boolean

Content that gives advice or instruction on how to commit a wrongdoing.

Example:

false

results.categories.illicit/violent

boolean

Illicit content that also references violence or procuring a weapon.

Example:

false

results.categories.self-harm

boolean

Content that promotes, encourages, or depicts acts of self-harm.

Example:

false

results.categories.self-harm/intent

boolean

Content where the speaker expresses that they are engaging or intend to engage in self-harm.

Example:

false

results.categories.self-harm/instructions

boolean

Content that provides instructions or advice on how to commit acts of self-harm.

Example:

false

results.categories.sexual

boolean

Content meant to arouse sexual excitement or that promotes sexual services.

Example:

false

results.categories.sexual/minors

boolean

Sexual content that includes an individual under 18 years old.

Example:

false

results.categories.violence

boolean

Content that depicts death, violence, or physical injury.

Example:

true

results.categories.violence/graphic

boolean

Content that depicts death, violence, or physical injury in graphic detail.

Example:

false

results.category_scores

object

Per-category confidence scores in [0, 1]. All 13 categories are always present, in OpenAI's order.

Hide child attributes

results.category_scores.harassment

number

Confidence score for harassment.

Example:

0.9412

results.category_scores.harassment/threatening

number

Confidence score for harassment/threatening.

Example:

0.9016

results.category_scores.hate

number

Confidence score for hate.

Example:

0.0231

results.category_scores.hate/threatening

number

Confidence score for hate/threatening.

Example:

0.0104

results.category_scores.illicit

number

Confidence score for illicit.

Example:

0.0057

results.category_scores.illicit/violent

number

Confidence score for illicit/violent.

Example:

0.0039

results.category_scores.self-harm

number

Confidence score for self-harm.

Example:

0.0021

results.category_scores.self-harm/intent

number

Confidence score for self-harm/intent.

Example:

0.0012

results.category_scores.self-harm/instructions

number

Confidence score for self-harm/instructions.

Example:

0.0008

results.category_scores.sexual

number

Confidence score for sexual.

Example:

0.0006

results.category_scores.sexual/minors

number

Confidence score for sexual/minors.

Example:

0.0002

results.category_scores.violence

number

Confidence score for violence.

Example:

0.8774

results.category_scores.violence/graphic

number

Confidence score for violence/graphic.

Example:

0.0311

results.category_applied_input_types

object

Which input modality triggered each category. This is a text-only model, so every category maps to ["text"].

Hide child attributes

results.category_applied_input_types.harassment

string[]

Example:

["text"]

results.category_applied_input_types.harassment/threatening

string[]

Example:

["text"]

results.category_applied_input_types.hate

string[]

Example:

["text"]

results.category_applied_input_types.hate/threatening

string[]

Example:

["text"]

results.category_applied_input_types.illicit

string[]

Example:

["text"]

results.category_applied_input_types.illicit/violent

string[]

Example:

["text"]

results.category_applied_input_types.self-harm

string[]

Example:

["text"]

results.category_applied_input_types.self-harm/intent

string[]

Example:

["text"]

results.category_applied_input_types.self-harm/instructions

string[]

Example:

["text"]

results.category_applied_input_types.sexual

string[]

Example:

["text"]

results.category_applied_input_types.sexual/minors

string[]

Example:

["text"]

results.category_applied_input_types.violence

string[]

Example:

["text"]

results.category_applied_input_types.violence/graphic

string[]

Example:

["text"]

Endpoints

ZeroClick

Batch API

llama-guard-4-12b

Authorizations

Headers

Body

Response