Mistral AI has released Shieldstral, an open-weight, multimodal safety classifier with 3 billion parameters. This model is designed for content moderation, allowing developers to set custom safety policies for different products and audiences. Shieldstral efficiently processes text, image, and embedded inputs, offering high flexibility and outperforming larger safety models with lower operational requirements. AI
IMPACT Provides developers with a flexible, efficient tool for content moderation across various modalities.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
