General-Purpose AI (GPAI) Model

An AI model trained on broad data that can perform a wide range of tasks and be integrated into many systems. The Act treats the model and the system built on it as two separate applicability questions: placing the GPAI model itself on the EU market is a provider act with its own obligations under Article 2(1)(a), and it does not by itself decide whether a downstream system built on that model is high-risk or who is its provider. Building on one can also trip the Article 25 provider line, which is why the upstream vendor relationship, what the model vendor owes and what you must verify yourself, is its own topic (see Topic 5.5).

An AI model trained on broad data and capable of many tasks, subject to a distinct set of provider obligations under the AI Act's GPAI chapter (Articles 51 onward); a substantial modification of such a model can carry GPAI provider duties on top of Article 4.

A broadly capable AI model, trained by a model provider, that can be adapted to many tasks and is typically reached through a paid API or self-hosted from open weights. The thing a wrap wraps. Under the EU AI Act, GPAI model providers carry their own obligations (applicable since 2 August 2025), which is part of what a deployer inherits exposure to. (see Topic 5.5)

Under the EU AI Act, a model trained on broad data that can perform a wide range of tasks and be integrated into many systems. GPAI providers carry specific duties, including publishing a training-content summary.

A model that can perform a wide range of tasks and be integrated into many downstream systems. Under the EU AI Act (Article 53, Regulation (EU) 2024/1689), GPAI model providers must, since 2 August 2025, supply documentation to downstream providers and publish a summary of training content.