Chapter 09 · Sustainability & AISustainability & AI
General-purpose AI (GPAI) model
Definition
Under the EU AI Act, an AI model trained with large-scale general data that displays significant generality and can competently perform a wide range of distinct tasks, integrable into many downstream systems.
References
This source provides part of the technical or institutional basis for the definition.
This source supports the explanation of how the term is applied, measured or governed in practice.
Overview
What it means
Chapter V of the Act assigns GPAI providers specific duties: technical documentation (including data provenance, training compute and known or estimated energy consumption), information to downstream providers, and copyright policy compliance. Models meeting a systemic-risk threshold — including cumulative training compute above 10^25 FLOPs — face additional evaluation and reporting duties.
How it is used
Foundation-model providers operationalise these duties through model documentation forms and codes of practice; downstream builders rely on the provided information for their own compliance. Limited exemptions exist for certain free and open-source models, but not where systemic risk is present.
Why it matters
GPAI rules regulate AI at the model layer rather than the application layer — a structural choice with global reach, since any provider serving the EU market falls in scope. The energy-documentation duty makes model footprints a compliance artefact for the first time.