Mistral AI's Shieldstral (open-weight) A 3B open-weights, policy-adaptive multimodal safety classifier that matches models up to 7x its size on text safety and sets a new state of the art on multimodal moderation. mistral.ai/news/shields...