{"rewrite":{"id":"r_8a817572ddf83f926ef4de39","clusterId":"c_dd4dfb6237c6bd6ba321f547","slug":"mistral-ai-releases-shieldstral-1-0-3b-moderation-model","model":"deepseek-v4-flash:free","headline":"Mistral AI Releases Shieldstral 1.0 3B Moderation Model","summary":"Mistral AI released Shieldstral 1.0 3B on August 4, 2026, a 3-billion-parameter AI model that judges whether text or images meet safety standards specified in natural language. It supports 12 languages, runs on a single GPU with 16GB VRAM, and is available under Apache License 2.0.","whyItMatters":"Shieldstral lets services define moderation criteria in natural language, removing the need for retraining when safety rules change.","webCardHtml":"\u003cp\u003eMistral AI\u0026#39;s Shieldstral 1.0 3B takes a different approach to content moderation. Instead of relying on predefined categories of dangerous content, it accepts safety criteria as questions, such as whether a text recommends violence or whether an image is safe for minors. The model reads the target and outputs probabilities for yes and no, which services can use to hide content or route it to human review.\u003c/p\u003e\u003cp\u003eThe model handles both text and images, and can also check whether an AI appropriately refused a dangerous request. In Mistral AI\u0026#39;s evaluations, Shieldstral matched larger models on new criteria without additional training and scored highest in image moderation among compared models. It supports 12 languages including English, Japanese, Chinese, and French, and runs on a single GPU with 16GB of VRAM. The weights are on Hugging Face under the Apache License 2.0.\u003c/p\u003e","blueskyPost":"Shieldstral 1.0 3B's natural-language rules let non-experts define safety criteria, but its 16GB VRAM requirement still ties it to a single GPU, limiting its appeal beyond well-resourced teams.","twitterPost":"Shieldstral 1.0 3B's natural-language rules lower the barrier for custom moderation, but the 16GB VRAM requirement still assumes a dedicated GPU.","threadsPost":"Shieldstral 1.0 3B trades the usual fixed safety taxonomy for natural-language rules, letting each team define its own standards. That flexibility is real, but the 16GB VRAM requirement means it is not a lightweight drop-in for every setup. Apache 2.0 licensing does lower the cost of trying it.","newsletterBlurb":"Mistral AI released Shieldstral 1.0 3B, a moderation model that evaluates text and images against safety criteria expressed in natural language. It supports 12 languages, runs on a single GPU with 16GB VRAM, and is available under Apache License 2.0.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20260805-mistral-shieldstral/\",\"title\":\"Mistral AI Releases Small Moderation AI 'Shieldstral' That Can Judge Text and Images with Natural Language Rules\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":4633,"outputTokens":589,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1786330438,"createdAt":"2026-08-10T02:36:46.000Z","publishedAt":"2026-08-10T02:46:44.000Z","updatedAt":"2026-08-10T02:36:46.000Z"},"cluster":{"id":"c_dd4dfb6237c6bd6ba321f547","canonicalTitle":"Mistral AIが小型モデレーションAI「Shieldstral」を公開、文章や画像を自然言語のルールで判定可能","representativeArticleId":"a_b1fea2eb89e5ef5b7191764e","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"Shieldstral 1.0 3B\"],\"studios\":[\"Mistral AI\"],\"people\":[],\"type\":\"announcement\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-08-05T05:50:00.000Z","lastSeenAt":"2026-08-05T05:50:00.000Z","updatedAt":"2026-08-10T02:46:44.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20260805-mistral-shieldstral/","title":"Mistral AIが小型モデレーションAI「Shieldstral」を公開、文章や画像を自然言語のルールで判定可能"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["Shieldstral 1.0 3B"],"studios":["Mistral AI"],"people":[],"type":"announcement","domain":"other","is_roundup":false},"keyFacts":null}