Mistral's open model Shieldstral matches much larger safety models at a fraction of the size

Mistral has released Shieldstral, a 3B parameter model designed to evaluate AI inputs and outputs for safety violations. By utilizing natural language prompts instead of rigid categories, the model allows operators to customize screening criteria at runtime. This approach enables a smaller system to achieve safety performance levels comparable to models seven times its size. The release offers developers a more efficient, adaptable tool for moderating content without relying on third-party constraints.