Model card
Also called system card
A document a lab publishes with a model, describing what it's for, how it was tested, its known limits and its risks.
The idea comes from a paper by Google researchers: a model should ship with a short report, like a nutrition label, saying what it was trained for, how it performs for different groups of people and where it shouldn't be used.
For today's big models these have grown into "system cards", long documents published at launch. They cover benchmark results, safety testing, attempts by outside experts to misuse the model, and the safeguards the company added. For a general reader they're the best single place to find what the maker itself says the model can't do well.
For example
Anthropic's system card for Claude Opus 5.5, published on 22 September 2026, is a long report of the model's test results, safety checks and known limits.
Put it to use
Where you'll meet model card in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.