Model card
A model card is a short document published with a machine learning model that explains what it is for, how it was trained and evaluated, where it performs poorly, and how it should and should not be used.
Also called: model documentation, README model card
By Divya Sharma · Checked Nov 15, 2026 · How we review
Mitchell et al. proposed model cards in 2018 to make model reporting more transparent. Their idea was that a released model should come with documentation of its intended uses and its measured performance across different conditions and groups of people, so it is less likely to be used where it does not work well.
What a model card contains
Typical sections are model details (who built it, version, architecture, model license), intended use and out-of-scope use, training data, evaluation results on public AI benchmarks, known limitations and biases, and safety notes. For LLMs you will often also find the context window, supported languages, prompt format and recommended settings such as temperature.
On Hugging Face, the model card is the repository's README: a Markdown file with extra metadata at the top. Labs releasing open-weight models usually publish one there.
How to use one
Read the intended-use and limitations sections before the benchmark table. Check what data the model was trained on (see AI training data) and whether the evaluations match your task and language. Note who ran the evaluations.
Limits
A model card is written by the model's maker. OWASP notes that model cards provide information but no guarantee of where a model really came from, so a convincing card on a copied or tampered model proves nothing. Verify the publisher and file hashes as well, especially for downloaded weights.
Related terms
- Open-weight model
- An open-weight model is a model whose trained weights are published for anyone to download and run, often without the training code or full training data, so it is not necessarily open source.
- AI training data
- AI training data is the collection of examples a model learns from, such as web text, code, images, transcripts or labelled pairs. Its size, quality, mix and licensing shape what the model can do and where it fails.
- Model license
- A model license is the legal terms under which a model's weights, and sometimes its code and data, may be used, changed and shared. Terms range from permissive open licences to non-commercial or custom restricted ones.
- AI benchmark
- An AI benchmark is a fixed, public set of tasks with an agreed scoring method, used to compare models on the same test, for example coding issues solved, exam questions answered or retrieval quality across many datasets.
- AI supply-chain attack
- An AI supply-chain attack compromises an AI application through something it depends on, such as a downloaded model, a LoRA adapter, a dataset, a Python package, an MCP server or an agent skill, rather than attacking the application directly.