Skip to content

Llama Guard

by Meta

Meta's open safety models that classify LLM prompts and responses, part of the Purple Llama project.

  • Self-hosted
  • Small business
  • Mid-size
  • Enterprise
  • Free tier

From Llama Guard's own website and public documents, updated 4 Nov 2026. Not a hands-on review: I add a verdict only to products I use myself.

What is Llama Guard?

Meta publishes Llama Guard, a family of input and output moderation models, in the Purple Llama repository. The repository describes Purple Llama as a set of tools to assess and improve LLM security. The models are fine-tuned from Llama 3.1 and 3.2 and detect content that breaks the MLCommons hazard taxonomy. The repo has folders for versions 2, 3 and 4.

Purple Llama also holds Prompt Guard for prompt injection and jailbreak attempts, Code Shield for insecure generated code, and CyberSec Eval benchmarks. The repository shows about 4.5k stars.

Licences vary. Benchmarks and Code Shield are MIT. Llama Guard models fall under the Llama community licences, which have their own terms of use.

Best forTeams that can host a classifier model and want a free safety filter

Llama Guard features

  • Moderation models

    Classify both prompts and responses against a hazard taxonomy.

  • MLCommons taxonomy

    Detects content that violates the MLCommons hazard categories.

  • Prompt Guard

    Targets prompt injection and jailbreak attempts.

  • Code Shield

    Filters insecure code produced by LLMs at inference time.

  • CyberSec Eval

    Benchmarks for insecure code and prompt injection risk.

  • Multiple versions

    Folders for Llama Guard 2, 3 and 4.

Llama Guard pricing

Starting price

Free (open weights, Llama community licence)

  • Pricing modelFlat fee
  • Free tierYes
  • Free trialNo

Current prices on the vendor's site · Checked 4 Nov 2026. Prices change often; the vendor's page wins.

Security and compliance

Certifications
None listed on a public trust page
EU data residency
Not documented

Where it fits (NIST CSF 2.0)

  1. Govern (not covered)
  2. Identify (not covered)
  3. Protect
  4. Detect (not covered)
  5. Respond (not covered)
  6. Recover (not covered)

Independent audits

No public independent audit found.

Good to know before you buy

  • Models use Llama community licences, which are not standard open-source licences; read the terms.
  • You must host and run the model yourself, which needs GPU or similar compute.

Llama Guard alternatives

Open-source Python framework with a validator hub for checking LLM inputs and outputs.

LLM Guard

Protect AI

Open-source scanner toolkit for LLM inputs and outputs; the repository is now archived.

Open-source NVIDIA toolkit for adding programmable input, dialog, retrieval, execution and output rails to LLM apps.

Securiti

Securiti, LLC

Data and AI security platform with privacy automation, DSAR, DSPM and LLM firewalls.

Llama Guard vs Guardrails AILlama Guard vs LLM GuardLlama Guard vs NeMo Guardrails

Frequently asked questions

What is Llama Guard?

Meta's open safety models that classify LLM prompts and responses, part of the Purple Llama project.

How much does Llama Guard cost?

Llama Guard starts at Free (open weights, Llama community licence), according to its pricing page (checked 4 Nov 2026).

Is there a free version of Llama Guard?

Yes. Llama Guard has a free plan or tier; check the pricing page for its limits.

What are the alternatives to Llama Guard?

Products in the same category include Guardrails AI, LLM Guard, NeMo Guardrails, Securiti.

Where is Llama Guard based?

The company behind Llama Guard is based in United States.

Have you tested Llama Guard?

Not yet. This page lists published facts only. When I test it, my verdict will appear at the top.

Sources (2), checked 4 Nov 2026
  1. github.com/meta-llama/PurpleLlama
  2. llama.com/llama-protections

Spotted something out of date? Tell me through the contact page and I will fix it.

By using this site, you agree to the Privacy Policy.