Mistral Launches Shieldstral, a 3B Open-Weights Safety Classifier for Text and Images

Mistral AI introduced Shieldstral, a 3-billion-parameter multimodal content moderation model that evaluates safety as plain-language Q&A instead of fixed categories, matching guard models up to 7x its size.