Examples
Worked examples
- Is an instance
A Hugging Face model card for a multilingual sentiment classifier listing per-language F1 scores and known failure modes.
- Is an instance
A model card accompanying a clinical-risk model documenting subgroup AUC across age, sex, and ethnicity.
Counter-examples
Looks similar, but isn't
- Not an instance
A model's training README with hyperparameters only.
- Not an instance
An academic-paper abstract.
Editorial commentary
A model card is a short, standardized document, published by the people who trained a specific model, reporting what the model is, how it was trained, its measured performance (including disaggregated performance across demographic and contextual slices), known limitations, and intended and out-of-scope uses. Introduced by Mitchell et al. (2019) to address the information asymmetry between model developers and everyone downstream of them — procurement staff, integrators, researchers reusing the model, and end users — a model card is meant to be read before a model is adopted, not audited after something goes wrong.
Model cards sit in the middle of a four-tier documentation hierarchy that this site treats as four genuinely distinct artefacts, each answering a different question: an algorithm card documents a method class before any specific model exists (pre-training, no trained instance); a model card documents one specific trained model, authored by the developer who trained it; a use card documents one specific deployment of that model in a particular application, authored by the deployer rather than the model’s developer; and a system card documents the full sociotechnical system a model is embedded in — the wrapper of moderation, retrieval, routing and user-facing controls around it. A model card is therefore developer-authored and model-scoped: it is not deployment-specific (that is a use card’s job) and it is not a description of the surrounding product (that is a system card’s job).
What a model card should report
- Model details: architecture, version, developer, licence, training-data cutoff.
- Intended use and out-of-scope use.
- Evaluation results, disaggregated by relevant subgroup where feasible.
- Known limitations, ethical considerations, and recommended mitigations.
Where it is used
Hugging Face Hub adopted a model-card README convention that has become the de facto standard for open-weight model releases; most major API-only providers (OpenAI, Anthropic, Google) publish an equivalent document, often folded into their system cards rather than issued as a separate model card.
References
- Mitchell et al., ‘Model Cards for Model Reporting’ (FAccT 2019)
- Hugging Face Hub Model Cards documentation
Also known as
ML model card
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="Model card"
vocab-term-identifier="https://casrai.org/dictionary/term/model-card" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/model-card",
"name": "Model card",
"identifier": "https://casrai.org/dictionary/term/model-card",
"description": "A short, structured document accompanying a machine-learning model that records its intended use, training data, evaluation methodology, performance characteristics across population sub-groups, and known limitations.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/ai-ml-research-outputs#set",
"url": "https://casrai.org/dictionary/term/model-card",
"sameAs": [
"ML model card"
],
"license": "https://creativecommons.org/licenses/by/4.0/",
"publisher": {
"@id": "https://casrai.org/#organization"
},
"author": {
"@id": "https://casrai.org/#editorial-team"
},
"datePublished": "2026-05-21T02:22:50",
"dateModified": "2026-08-22T15:54:47",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
