Examples
Worked examples
- Is an instance
OLMo from AI2 with full training data, code, intermediate checkpoints, and weights released.
- Is an instance
Pythia model suite with full reproducible training pipeline.
Counter-examples
Looks similar, but isn't
- Not an instance
An open-weights model with undisclosed training data (not OSAID-compliant).
- Not an instance
An API-only proprietary model.
Editorial commentary
The Open Source AI Definition (OSAID v1.0, published by the Open Source Initiative in October 2024) is the first formal, versioned criteria set for what “open source” means applied to an AI system, rather than to conventional software. It requires that a genuinely open-source AI release give downstream users enough information about the training data, the code used to train and build the system, and the model’s architecture and parameters to substantially recreate an equivalent system — a materially higher bar than publishing a downloadable weights file alone.
This directly contests how many widely used models are marketed. Under OSAID’s data-disclosure requirement, models commonly labelled “open” by their developers — Llama, Gemma, Qwen among them — do not qualify, because their training data is not disclosed to the required level; OSI’s own guidance places these in the separate open-weights category instead. The RAIL Initiative, which issues the OpenRAIL family of model licences, states in its own FAQ that OpenRAIL licences are explicitly not open-source licences under the OSI definition, because they carry use-based restrictions that also bind anything derived from the model — a condition OSAID and traditional open-source licensing both reject as a matter of principle. Some commentary has gone further and characterised “open source” framing applied to weights-only releases with restrictive commercial terms as open-washing; that characterisation is contested and attributed to specific critics rather than treated here as settled fact.
Why this dispute matters, not just how to define it
For a research office evaluating a model for adoption, “open source” is sometimes used as shorthand for “no licensing risk” or “freely reproducible” — neither of which follows automatically from a weights-only release. OSAID gives procurement and research-integrity reviewers a citable, versioned criteria set to check a specific release against, rather than relying on the label a vendor’s marketing applies to it. The definition is expected to be revised over time; verify against the current version at opensource.org/ai rather than assuming this description stays current indefinitely.
References
- Open Source Initiative, ‘The Open Source AI Definition’ v1.0 (October 2024), opensource.org/ai
- RAIL Initiative FAQ, licenses.ai/faq-2
Also known as
OSAID-compliant model · open-source AI (OSI)
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="Open-source model (criteria)"
vocab-term-identifier="https://casrai.org/dictionary/term/open-source-model-criteria" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/open-source-model-criteria",
"name": "Open-source model (criteria)",
"identifier": "https://casrai.org/dictionary/term/open-source-model-criteria",
"description": "A model meeting the criteria articulated by the Open Source Initiative's Open Source AI Definition: open data information, open code, and open weights, with each released under terms compatible with the OSI's freedoms to use, study, modify, and share.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/ai-ml-research-outputs#set",
"url": "https://casrai.org/dictionary/term/open-source-model-criteria",
"sameAs": [
"OSAID-compliant model",
"open-source AI (OSI)"
],
"license": "https://creativecommons.org/licenses/by/4.0/",
"publisher": {
"@id": "https://casrai.org/#organization"
},
"author": {
"@id": "https://casrai.org/#editorial-team"
},
"datePublished": "2026-05-21T02:22:51",
"dateModified": "2026-08-22T15:54:50",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
