Model card

Definition

A standardized documentation format that describes a model's intended use, training data, performance metrics, limitations, and ethical considerations. Model cards promote transparency and help users make informed decisions about model selection.

Why it matters

Model cards are becoming the nutrition labels of AI. As models proliferate and regulation tightens, standardized documentation is essential for responsible procurement and deployment. A good model card tells you what a model was trained on, where it excels, where it fails, and what biases it exhibits. Without model cards, enterprises are deploying black boxes with no way to assess fitness for their use case. The EU AI Act requires documentation similar to model cards for high-risk AI systems. Hugging Face has made model cards standard practice for open-weight models, and enterprise buyers are increasingly requiring them from commercial vendors.

In practice

Google introduced the model card concept in 2018 with a template that included model details, intended use, performance metrics, and ethical considerations. Hugging Face made model cards mandatory for all models on its platform, creating the largest corpus of model documentation. Anthropic publishes detailed model cards for each Claude version, including benchmark results, safety evaluations, and known limitations. Meta's Llama model cards include training data composition, red teaming results, and responsible use guidelines. In practice, model cards help enterprise procurement teams evaluate model suitability without running expensive proof-of-concepts for every candidate.