Mistral AI(@MistralAI)
Shieldstral is available under Apache 2.0. Try it: https://t.co/Hqlo3o7t6M
8.5内容质量
TL;DR · AI 摘要
Mistral AI发布Shieldstral，一个3B参数的开放权重内容安全模型，支持多模态审核并可在设备部署。
核心要点
- Shieldstral是3B参数模型，基于Apache 2.0许可证开放
- 支持文本和图像审核，使用自然语言输入返回校准评分
- 可部署于单块16GB NVIDIA GPU，企业可自定义安全策略
结构提纲
按章节快速跳转。
- §模型发布
Mistral AI宣布推出Shieldstral模型并开放Apache 2.0许可证
- ·技术特性
支持文本和图像多模态审核，使用自然语言输入返回校准评分
- ›部署能力
可在单块16GB NVIDIA GPU上运行，支持设备端部署
- ·企业应用
提供定制化安全策略配置功能，满足企业合规需求
思维导图
用一张图看清主题之间的关系。
查看大纲文本（无障碍 / 无 JS 友好）
- Shieldstral模型
- 模型特性
- 3B参数规模
- Apache 2.0许可证
- 技术优势
- 多模态审核
- 自然语言接口
- GPU高效运行
- 应用场景
- 企业内容安全
- 设备端部署
金句 / Highlights
值得收藏与分享的关键句。
Shieldstral是3B参数的开放权重模型，可在设备上部署
支持文本和图像审核，使用自然语言输入返回校准评分
运行于单块16GB NVIDIA GPU，效率达到行业领先水平
#内容安全#AI模型#多模态审核#Apache 2.0
打开原文Mistral AI on X: "Shieldstral is available under Apache 2.0. Try it: https://t.co/Hqlo3o7t6M" / X
[](https://x.com/)
Mistral AI on X: "Shieldstral is available under Apache 2.0. Try it: https://t.co/Hqlo3o7t6M"
-  Mistral AI @MistralAI Aug 4 🛡️Introducing Shieldstral, Mistral’s 3B open-weights model for content safety that can be deployed on-device 🧵 mistral.ai/news/shieldstr…  [](https://x.com/MistralAI/status/2084684735725379637/photo/1) 208 252 2.8K 496K
-  Mistral AI @MistralAI Aug 4 State-of-the-art on multimodal moderation, Shieldstral boasts industry-leading efficiency, running on a single 16GB NVIDIA GPU and gives enterprises customized control of what’s deemed safe. 5 5 139 24K
-  Mistral AI @MistralAI Aug 4 The model takes moderation policy as a plain-language question and returns a calibrated score. Text and images — one interface. Read the full technical report here: arxiv.org/abs/2607.25857  [](https://x.com/MistralAI/status/2084684738892083270/photo/1) 8 12 127 45K
-  Mistral AI @MistralAI Shieldstral is available under Apache 2.0. Try it: From huggingface.co 8:11 PM · Aug 4, 202616K Views 5 7 79 25
-  Krish Dasgupta @officialKrishD Aug 4 Releasing this in few hours. This will change how people have perceived clustering and classification and reasoning when it comes to guardrails and scenarios. This is how it looks , @MistralAI - The shieldMistral on NetworkX graph, reasoned through mistral-MLX-4bit running Apple Show more  [](https://x.com/officialKrishD/status/2084791078394475003/photo/1) 1 181
-  Strategize Labs @StrategizeLabs Aug 6 Open, local guardrails are the right default for regulated environments. That’s where enterprises get a grip on data security. 118
-  makassar_iinfo @makassar_infoo Aug 5 Releasing a compact multimodal moderation model under Apache 2.0 is a strong move for on-device safety. The practical advantage will be seeing how well the plain-language policy interface handles domain-specific rules without requiring every deployment to build a separate Show more 262
Log in or sign up for X
See what’s happening and join the conversation
Continue with phoneContinue with Apple Continue with Google
or