Introducing Shieldstral
Mistral AI, Tuesday, August 4th, 2026
Mistral AI releases Shieldstral, a 3B open-weights multimodal safety classifier outperforming models 7x larger.
Mistral AI has introduced Shieldstral, a safety classifier released with open weights.
The model has 3 billion parameters and handles multimodal input rather than text alone. Mistral reports it outperforms models up to seven times its size on safety classification tasks.
The compact size makes it practical to run as a filtering layer without a disproportionate inference cost.