by Meta AI

Llama Guard 4 12B.

multimodal open weights laptop+ 12B params 164K ctx
Cheapest input
$0.18/M
on DeepInfra
Cheapest output
$0.18/M
on DeepInfra
Smallest GPU
1× Nvidia GeForce RTX 3060
$0.038/hr
Description

About Llama Guard 4 12B.

Llama Guard 4 12B is Meta's safety classifier, not a general assistant. It was pruned from Llama 4 Scout and tuned to label prompts and responses as safe or unsafe, naming the categories broken. It reads text and multiple images. Meta tested it mostly with about three images, warns that prompt injection could bypass it, and says hazards needing current facts, such as elections, need more complex systems. Weights are gated under the Llama 4 Community License.

Context window

How much it can remember.

164K tokens ≈ 122,880 English words
4K 32K 128K 1M
Capabilities

What it can do.

· Vision input
· Audio input
· Video input
· Function calling
· Tool use
· JSON mode
✓ Streaming
· Fine-tuning