Mistral releases self-hostable 3B moderation model Shieldstral 1.0

Mistral has released Shieldstral 1.0, a 3-billion-parameter model designed for self-hosted text and image content moderation. Available under an Apache 2.0 license with full weights, the model can run on a single 16GB GPU and integrates with common inference stacks. Its key feature is "policy-adaptive" moderation, allowing users to define moderation rules in natural language prompts without retraining. While Mistral reports strong benchmark scores, these are not yet independently verified, and the model is specialized for moderation rather than general-purpose tasks. AI

IMPACT Enables self-hosted, privacy-preserving content moderation for smaller deployments and homelabs.

RANK_REASON Release of an open-weight, specialized model with detailed technical specifications and benchmark claims, but not from a tier-1 frontier lab. [lever_c_demoted from research: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• 16GB GPU
• Apache 2.0
• Axolotl
• ByteDance
• DeepSeek V4-Pro
• HarmBench
• Hugging Face
• llama.cpp
• Ministral-3-3B-Base-2512
• Seed 2.1 Turbo
• SGLang
• Shieldstral 1.0
• ToxicChat
• Transformers
• VLGuard
• vLLM

• model release
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →