Mistral has released Shieldstral, a new 3B open-weights multimodal safety classifier designed for content moderation. The model is specifically built to evaluate and filter multimodal content for safety. Despite its compact size, Shieldstral outperforms competing moderation models that are up to seven times larger.