In plain words
A model card accompanies a trained AI model and explains its intended use, evaluation methods, performance, and relevant limitations. It helps people judge whether the model fits their needs.
A closer look
Useful reporting describes the conditions under which a model was tested and how performance varies across relevant groups or situations. A single overall score can hide important differences.
Model cards vary in format and completeness. Read them alongside your own evaluations, especially when your use case differs from the conditions described.
In practice
Before choosing an image classifier, a team checks its model card for intended uses and performance across lighting conditions and demographic groups.
A useful distinction
A model card is documentation, not a safety certificate or an independent audit. It does not guarantee that the model will perform well in a new setting.