Title: Llama Guard 3 8B vs Llama Guard 4 12B: Param%C3%A8tres %E2%80%94 LLM EU
Canonical: https://llmeu.com/fr/compare/llama-guard-3-8b-vs-llama-guard-4-12b/
Markdown mirror: https://llmeu.com/fr/compare/llama-guard-3-8b-vs-llama-guard-4-12b.md
Machine-readable: https://llmeu.com/llms.txt (index) - https://llmeu.com/llms-full.txt (full catalogue) - https://llmeu.com/llms-catalogue.json

# Llama Guard 3 8B vs Llama Guard 4 12B

Placez deux %C3%A0 quatre mod%C3%A8les c%C3%B4te %C3%A0 c%C3%B4te sur les champs qui d%C3%A9cident un d%C3%A9ploiement : contexte, prix, r%C3%A9gion, licence, langues, h%C3%A9bergement et ce qui est v%C3%A9rifi%C3%A9.

Lorsque deux mod%C3%A8les diff%C3%A8rent, ce tableau montre la diff%C3%A9rence et s'arr%C3%AAte l%C3%A0. LLM EU ne publie aucun badge de gagnant sans score document%C3%A9 pour cette t%C3%A2che, et une valeur manquante reste manquante au lieu de devenir un z%C3%A9ro.

## Quand l'utiliser

### [Llama Guard 3 8B](/fr/models/llama-guard-3-8b/)

#### Points forts

- open weights are published on Hugging Face and can be downloaded without a vendor agreement
- at 8B parameters it fits on a single accelerator
- the model page lists en
- instruction-tuned checkpoint with a chat template in the repository

#### Limites

- catalogue only %E2%80%94 LLM EU does not run this model, and nothing on this card implies an LLM EU endpoint
- no evaluation has been run by LLM EU
- the published configuration records no context window, so the card carries none
- at 8B parameters, the weights fit on a single modest accelerator
- a classifier, not a chat model: it labels prompts and responses against a safety taxonomy instead of answering them

#### T%C3%A2ches

[Donn%C3%A9es sensibles](/fr/tasks/private/), [Support client](/fr/tasks/customer-support/)

### [Llama Guard 4 12B](/fr/models/llama-guard-4-12b/)

#### Points forts

- open weights are published on Hugging Face and can be downloaded without a vendor agreement
- at 12B parameters it fits on a single accelerator
- the model page lists en
- instruction-tuned checkpoint with a chat template in the repository

#### Limites

- catalogue only %E2%80%94 LLM EU does not run this model, and nothing on this card implies an LLM EU endpoint
- no evaluation has been run by LLM EU
- the llama4 (other) licence is a custom agreement whose commercial terms were not established from its published text, so commercial serving is not permitted until a human clears it; the router must not route to it
- the published configuration records no context window, so the card carries none
- at 12B parameters, self-hosting needs a single accelerator with enough memory for the weights
- vision checkpoints are heavier to serve than their parameter count alone suggests, because the image encoder and pre-processing run alongside the language model
- a classifier, not a chat model: it labels prompts and responses against a safety taxonomy instead of answering them

#### T%C3%A2ches

[Donn%C3%A9es sensibles](/fr/tasks/private/), [Vision](/fr/tasks/vision/), [Support client](/fr/tasks/customer-support/)

**T%C3%A2ches:** [Donn%C3%A9es sensibles](/fr/tasks/private/), [Support client](/fr/tasks/customer-support/)

## Comparer des mod%C3%A8les

- [Aya Expanse 8B vs Llama Guard 3 8B](/fr/compare/aya-expanse-8b-vs-llama-guard-3-8b/)
- [Gemma 3 12B It vs Llama Guard 4 12B](/fr/compare/gemma-3-12b-it-vs-llama-guard-4-12b/)
- [Gemma 4 12B It vs Llama Guard 4 12B](/fr/compare/gemma-4-12b-it-vs-llama-guard-4-12b/)
- [Gemma 4 E4B IT vs Llama Guard 4 12B](/fr/compare/gemma-4-e4b-it-vs-llama-guard-4-12b/)
- [GLM 4.1V 9B Thinking vs Llama Guard 4 12B](/fr/compare/glm-4-1v-9b-thinking-vs-llama-guard-4-12b/)
- [GLM 4.6V Flash vs Llama Guard 4 12B](/fr/compare/glm-4-6v-flash-vs-llama-guard-4-12b/)