Meta: Llama Guard 4 12B

meta-llama/llama-guard-4-12b · MetaNot yet established

Declared family: Unknown
Evidence confidence: Unknown
Declared tokenizer tag: Other
Entered catalogue: 30 April 2025
Last tested: 21 August 2026
Measurement: Measured

Identity statement

Meta: Llama Guard 4 12B has a clean measured fingerprint that currently sits alone: no other measured model shares its signature.

Evidence stack

Signals of different kinds, weighed together

Signal	Result	Reference
Tokenizer signature	Clean measurement; no exact match yet	tk_985608d3
API surface	No other entry shares this exact contract	api_53fdcc38
Context and output	1,048,576 context · 16,384 max output	declared
Reasoning contract	None declared	declared
Serving providers observed	Together	measured 21 August 2026

Closest measured models

Agreement across the mutually clean test strings

Meta: Llama 4 Scoutmeta-llama/llama-4-scout30/50 · 60%Meta: Llama 4 Maverickmeta-llama/llama-4-maverick30/50 · 60%Meta: Muse Glimmer 30Bmeta/muse-glimmer-30b28/50 · 56%Reka Edgerekaai/reka-edge19/50 · 38%OpenAI: GPT-5 Nanoopenai/gpt-5-nano17/50 · 34%Cohere: North Mini Code (free)cohere/north-mini-code:free17/50 · 34%

Click a row to open the full comparison.

The measured fingerprint

Marginal prompt-token cost of each test string, grouped by script

All 50 rows

English and whitespace

en-prose14

en-long16

spaces-203

spaces-603

tabs-203

newlines-204

mixed-ws6

Digits

digits-94

digits-125

digits-3011

digits-sep8

float-long10

CJK

zh-common11

zh-long14

zh-rare19

ja-kana9

ja-kanji11

ko9

Other scripts

ru10

ar11

he10

hi12

th8

el14

Emoji

emoji-basic11

emoji-skin20

emoji-zwj-family12

emoji-zwj-x334

emoji-flags17

emoji-prof20

Rare Unicode

math24

boxdraw22

combining10

cjk-ext-b17

surrogates37

zalgo31

rtl-mix6

Code

py-code25

py-indent15

json20

html17

regex43

camel6

snake6

Repetition and encodings

rare-word-x530

repeat-tok40

base6435

hex11

url17

uuid28

An x marks a row excluded as corrupt (caching or provider interference detected during measurement). Overhead subtracted: 205 prompt tokens.

Declared record

What the catalogue claims about this model

Llama Guard 4 is a Llama 4 Scout-derived multimodal pretrained model, fine-tuned for content safety classification. Similar to previous versions, it can be used to classify content in both LLM...

Modalities: text+image->text
Declared tokenizer: Other
Prompt price: $0.18 / M tokens
Completion price: $0.18 / M tokens

Supported parameters

frequency_penaltylogit_biasmax_tokensmin_ppresence_penaltyrepetition_penaltyresponse_formatseedstoptemperaturetop_ktop_p

defaults: {}

Catalogue entry Weights on Hugging Face

History

Every observation, kept as taken

30 April 2025Enters the OpenRouter catalogue with no declared family.
21 August 2026Fingerprinted in the YFarmX catalogue sweep · 50 of 50 strings measured clean.

Cite this page as the evidence record for Meta: Llama Guard 4 12B: the URL is stable, measurements are dated, and revisions append to the history above. Method: tokenizer fingerprinting, v1, 50 strings. The raw data behind every figure is on the data page.