Hacker News · item history
Mistral's Shieldstral: 3B open-weights model for multimodal moderation
8
first seen points
481
peak points
481
latest points
35
observations
Posted on X
- Mistral shipped a 3B moderation model you can run yourself. The use case: you want to reject toxic inputs *before* they hit your main LLM, not hope the model refuses them. Cheaper and faster than a safety prompt on every call.