Mistral Releases Shieldstral, a 3B Open-Weights Model for Multimodal Moderation

Mistral released Shieldstral, a 3-billion-parameter open-weights model built to classify harmful content across both text and images. At that size it is meant to run as a cheap guardrail in front of larger models rather than as a general assistant. Open weights matter here because moderation policies differ by jurisdiction and by product, and teams that can fine-tune the classifier can encode their own policy instead of accepting a vendor default.