The European Commission (the Commission) recently launched a consultation on non-binding guidelines for providers of general-purpose AI models (GPAI) models under the EU AI Act: Commission seeks input to clarify rules for general-purpose AI models | Shaping Europe’s digital future.

The guidelines “will give detailed explanations on questions such as ‘what is a general-purpose AI model’, ‘which entities are providers in various constellations’, and ‘which actions constitute a placing on the market’. The guidelines will also lay out how the AI Office will provide support to facilitate compliance. Finally, they will explain how signing the Code of Practice – if approved by the AI Office and the AI Board – may reduce administrative burden for providers and serve as a benchmark for regulatory compliance. In this respect the guidelines will complement the Code of Practice on General-Purpose AI, which is currently being finalised.”

The Commission says the new guidelines will help GPAI model providers comply with the EU AI Act by clarifying key concepts, such as:

• The definition of “general-purpose AI model"; to establish whether a model displays "significant generality and is capable of competently performing a wide range of distinct tasks". The preliminary view is that training compute is "an imperfect proxy for generality and capabilities but currently the best metric for legal certainty" with a presumption that a model that can generate text and/or images with a training compute of more than 10^22 FLOP (the measure of floating-point operations per second, measuring computer performance) will typically qualify as general-purpose. 
• When does a downstream modifier become a provider; the thresholds in terms of computational resources used for the modification, that if met, mean the downstream modifier should be presumed to be the provider of the GPAI model or GPAI model with systemic risk, and subject to the relevant obligations in the EU AI Act. 
• How to estimate the computational resources used for training a GPAI model.

Beyond the conceptual clarifications, the guidelines are expected to clarify how the AI Office will work with providers who must comply with the general-purpose AI rules, to support them in their compliance. It is intended that the guidelines on GPAI will complement the General-Purpose AI Code of Practice (AI Code of Practice) which will set out the commitments to which providers of GPAI models may adhere to ensure compliance with their obligations under the EU AI Act. These clarifications are intended to facilitate compliance with the EU AI Act in general, and to serve as a benchmark for regulatory compliance.

Feedback from stakeholders (including from GPAI model providers, providers of GPAI systems, public authorities, civil society, and academia) is required by 22 May 2025. Both the guidelines and the final AI Code of Practice are expected to be published ahead of August 2025. It is further expected that the Commission will launch a similar targeted consultation on the classification of AI systems as “high risk”.