General-Purpose AI Model Documentation

How to use this document

• What it is for. It provides a reference structure for the provider of a general-purpose AI model to organize the documentation under arts. 53 to 55 of Regulation (EU) 2024/1689 (the "AI Act"): the model's technical documentation (Annex XI), information for providers that integrate it into their systems (Annex XII), the copyright policy and preparation of the public summary of training content. It includes the additional obligations for models with systemic risk.
• Public summary of training content. This document does not reproduce the AI Office's official template for that summary. Download and use the current version published by the Commission and its explanatory notice; Part D only helps you gather the internal information needed to complete it.
• Who should own it. The model owner (research or model engineering), with legal (intellectual property and regulatory), data governance and security. Enter the function in [RESPONSIBLE FUNCTION] and the person in [OWNER].
• Decisions to take first. Confirm with your advisor: whether the model is general-purpose and whether the organization is its provider or only modifies or integrates it (section 2); whether it can rely on the open-source exception (section 2.3); whether the model has systemic risk (section 2.4); and whether the organization signs up to the general-purpose AI code of practice.
• Code of practice. If your organization signs up to the code of practice, check whether it provides its own forms (for example, for model documentation) and use them as the primary format; this structure can serve as an internal index and checklist.
• Dates. The application dates, the transitional regime for models placed on the market before the obligations applied and the date from which the AI Office can impose fines are shown in "Related regulatory references." Confirm them with your advisor.
• Fields to complete. Replace [AI SYSTEM] with the model name, [LEGAL ENTITY NAME], [OWNER], [AI GOVERNANCE EMAIL] and the [COMPLETE] fields. Refer to long documents by code and version instead of copying them.
• Review and approval. Have your legal counsel review sections 2, 5 and 6 in particular. Approve each version and record it in "Version history."

General-Purpose AI Model Documentation

1. Control data

Field	Value
Model	[AI SYSTEM]
Version or variants covered	[COMPLETE]
Provider	[LEGAL ENTITY NAME]; [COMPLETE: address and contact]
EU authorized representative (if the provider is not established in the EU)	[COMPLETE]
EU release date	[DATE]
Distribution methods	[COMPLETE: API, weights download, integration into own products, other]
License	[COMPLETE]
Model with systemic risk?	[COMPLETE: yes, no, under analysis]
Signed up to the code of practice?	[COMPLETE]
Documentation owner	[OWNER]
Contact for integrators and authorities	[AI GOVERNANCE EMAIL]
File version and date	[COMPLETE]; [DATE]

2. Scope and classification

2.1 Is it a general-purpose AI model?

Question	Answer	Evidence
Does the model display significant generality and competently perform a wide range of distinct tasks?	[COMPLETE]	[COMPLETE]
Can it be integrated into a variety of downstream systems or applications?	[COMPLETE]	[COMPLETE]
Is it used only for research, development or prototyping before being placed on the market?	[COMPLETE]	[COMPLETE]
Indicative criteria from the Commission guidelines applied	[COMPLETE: confirm with your advisor the current criteria]	[COMPLETE]
Conclusion	[COMPLETE]	[COMPLETE]

2.2 The organization's role

Question	Answer
Does it develop the model, or have it developed, and place it on the market under its name or trademark?	[COMPLETE]
Does it modify (for example, through fine-tuning) a third party's model? To what extent?	[COMPLETE: confirm with your advisor whether the modification makes it the provider of the modified model and for which obligations]
Does it only integrate a third party's model into its own system? (if so, the obligations in this document fall on the model provider, and the organization needs the Part B information)	[COMPLETE]
Resulting role	[COMPLETE]

2.3 Open-source exception

The technical documentation (Part A) and integrator information (Part B) obligations do not apply to models released under a free and open-source license that allows access, use, modification and distribution, where their parameters (including weights), information on their architecture and on their use are publicly available. The exception does not apply to models with systemic risk and does not exempt from the copyright policy (Part C) or the public summary (Part D).

Condition	Met?	Evidence
Free and open-source license allowing access, use, modification and distribution	[COMPLETE]	[COMPLETE]
Parameters, including weights, public	[COMPLETE]	[COMPLETE]
Architecture information public	[COMPLETE]	[COMPLETE]
Model usage information public	[COMPLETE]	[COMPLETE]
The model does not have systemic risk	[COMPLETE]	[COMPLETE]

2.4 Systemic risk

2.4.1. A model is presumed to have high-impact capabilities when the cumulative amount of computation used for its training, measured in floating-point operations, is greater than 10^25. The Commission may amend this threshold and may also designate models based on other criteria (arts. 51 and 52).

2.4.2. Estimated cumulative training compute: [COMPLETE], using the estimation method: [COMPLETE].

2.4.3. If the threshold is met, or it becomes known that it will be met, [ORGANIZATION NAME] notifies the Commission without delay and in any event within two weeks, with the information showing this. With the notification it may present sufficiently substantiated arguments that, exceptionally, the model does not present systemic risks. Notification date: [DATE]. Owner: [OWNER].