MAMistral AI/ Shieldstral
Shieldstral 1.0 3B
mistral/shieldstral-1.0-3b
A compact policy-adaptive multimodal safety classifier for text and images. It evaluates content against policies written in natural language and produces a continuous safety score.
Model information
Released
2026-08-04
Context window
32K
Model parameters
3B
Architecture
Ministral 3 3B with a Pixtral vision encoder
Weights
Open
Modalities
Input
ImageText
Output
Text
Identity
- Version
- 1.0
- License
- Apache-2.0
- Aliases
- Shieldstral-1.0-3B, mistralai/Shieldstral-1.0-3B
Capabilities
Open weight