Mistral launched Shieldstral, a 3B Apache 2.0 open-weights content safety model optimized for on-device deployment on a single 16GB GPU with natural-language moderation policy evaluation.

Key Takeaways

  • ✓Fully open-source under Apache 2.0, 3B parameters optimized for single 16GB GPU or edge inference;
  • ✓Evaluates arbitrary plain-language safety policies with calibrated risk scores across text and vision;
  • ✓Weights available on Hugging Face with accompanying technical whitepaper on arXiv.