Examples
Worked examples
- Is an instance
Reporting 'gpt-4-0613 (OpenAI, accessed 2024-03-15)' rather than just 'GPT-4'
Counter-examples
Looks similar, but isn't
- Not an instance
Reporting only 'ChatGPT' without date or version is insufficient for reproducibility
Editorial commentary
Model versioning identifies a specific revision of an AI model precisely enough to distinguish it from earlier or later revisions that may behave differently on the same input — by name, version number, release date, git-style commit hash, or content hash of the weights file itself. It answers “which exact model was this?” rather than “how was this model produced?”.
Why closed-API models make this hard
Open-weight models version cleanly: a downloaded weights file has a fixed hash and never silently changes. Closed, API-served models do not — a provider can update the model served under a stable-sounding name (a “latest” alias, for example) without publishing a new version number, so two calls to the same endpoint months apart may not be running the same model at all. This is a genuine reproducibility problem for research: a result obtained by prompting an API-served model is only reproducible if the exact snapshot used was recorded (a dated snapshot identifier where the vendor offers one, or at minimum the query date, as a fallback).
How this differs from lineage and checkpoints
Versioning is an identification practice, not a provenance record: it labels a revision, it does not describe where that revision came from. Model lineage answers the derivation question; versioning answers the identification question. A checkpoint is an internal training-time artefact that may never receive a public version label at all — versioning is usually reserved for artefacts intended for external release or use.
Practical recommendation
When citing or reporting an AI model’s output in a manuscript or protocol, record the exact version identifier and query date, not just the model family name — the same discipline already required for software and dataset citation.
References
Also known as
Model snapshot · Model release
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="Model versioning"
vocab-term-identifier="https://casrai.org/dictionary/term/model-versioning" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/model-versioning",
"name": "Model versioning",
"identifier": "https://casrai.org/dictionary/term/model-versioning",
"description": "The practice of identifying a specific revision of an AI model by name, version number, release date, or content hash, sufficient to uniquely distinguish it from earlier or later revisions that may behave differently on the same input.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/genai-disclosure#set",
"url": "https://casrai.org/dictionary/term/model-versioning",
"sameAs": [
"Model snapshot",
"Model release"
],
"license": "https://creativecommons.org/licenses/by/4.0/",
"publisher": {
"@id": "https://casrai.org/#organization"
},
"author": {
"@id": "https://casrai.org/#editorial-team"
},
"datePublished": "2026-05-21T01:57:42",
"dateModified": "2026-08-22T15:43:51",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
