AIght_
ToolsLearnFieldsUniverseSignalHumanAbout
Take the quiz
← All concepts

Concept

Model Cards

The documentation labs publish when they release a model — and what they leave out.

Mankaran Singh·Updated May 17, 2026

Where this idea lives

PREREQUISITESTOOLS THAT SHOW ITModel CardsEvalsEvals — How you measure whether a model is good at the thing you actually care about.AI Safety & AlignmentAI Safety & Alignment — The problem of building AI that reliably does what you actually wanted — not what you literally asked forHow AI Models Are TrainedHow AI Models Are Trained — From random noise to a model that can reason — the actual pipelineClaudeClaudeChatGPTChatGPTGeminiGeminiCommon misconception: Model cards are objective spec sheets.Common misconception: Disclosed limitations are the actual limitations.Common misconception: Two models with similar cards behave similarly.
prereqsrelatedtoolsmisconceptions
shows up in:Software EngineeringSocial Work & Public PolicyEducation & TeachingLaw & Legal
You might think:Model cards are objective spec sheets.Disclosed limitations are the actual limitations.Two models with similar cards behave similarly.

Common misconception

“Model cards are objective spec sheets.”

They're marketing artefacts shaped by what the lab chose to disclose, on benchmarks the lab chose to run, with limitations described in language the lab's policy team approved. The numbers are usually honest. The framing is rarely neutral. Two model cards with similar headline benchmarks can describe models that behave very differently in your task.

A model card is the document a lab publishes alongside a model release. The format was proposed in 2018 (Mitchell et al.) and has become standard practice — though the rigour varies wildly.

What a good card includes

  • Identity. Architecture, parameter count, training data description, training compute.
  • Intended use. Where the model is meant to be deployed.
  • Limitations. Failure modes the lab has documented.
  • Evaluations. Benchmark results — usually selective.
  • Ethical considerations. Bias evaluations, safety testing.
  • Training data. Sources, filtering criteria, known issues.

What's usually missing

  • Full training data. "Web text and books" is the typical disclosure. The specifics — which books, which web sources, what was excluded — are commercially sensitive.
  • Negative results. Benchmarks the model did poorly on rarely appear.
  • Behavioural quirks. The model's idiosyncratic voice, refusal patterns, sycophancy tendencies — usually discovered post-launch by users.

How to read one

Search for the bullet points the lab spent the most words on — those are usually where their genuine value lies. Then run your own eval.

What to read next

Evals are how you verify the card's claims on your task. Alignment covers the ethical-considerations section.

← Back to all conceptsBrowse tools →
beginner
Read time4 min read
UpdatedMay 2026
Sources4

Read next

  1. Evals →
  2. AI Safety & Alignment →
  3. How AI Models Are Trained →