GPAI stands for General-Purpose AI Model and refers to a category of artificial intelligence models that exhibit broad applicability and can competently perform a wide range of different tasks. Unlike narrow AI models designed for specific, single-use applications, GPAI models are capable of generalizing across domains and can be integrated into many different downstream systems or applications.
Core Characteristics:
A General-Purpose AI Model is defined by its versatility and adaptability. It is not tied to one fixed task but can execute various functions such as text generation, image creation, audio processing, question answering, summarization, translation, and more. Well-known examples of GPAI models include ChatGPT, DALL·E, and Midjourney, which can operate in multiple modalities and contexts.
GPAI models are based on foundation models, meaning their classification as GPAI depends not on a specific application but on the functionality and generality of the underlying model. These models can be fine-tuned or adapted for a wide range of applications, making them powerful but also more challenging to regulate and monitor.
Regulatory Context (AI Act – EU):
Due to their broad capabilities, GPAI models introduce unique risks—such as misuse, misinformation, or opaque behavior—which is why the EU AI Act includes specific rules for them:
-
Transparency Requirements: Developers must document the use of copyrighted training data and clearly label AI-generated content.
-
Technical Documentation: GPAI providers are obligated to submit comprehensive technical information about the model’s development and intended use.
-
Risk Management: For powerful GPAI models that could pose systemic risks, additional obligations apply. These include:
-
Incident reporting
-
Red teaming and adversarial testing
-
Ongoing monitoring of serious malfunctions
-
Cybersecurity and governance safeguards
-
The EU AI Office uses the training compute budget (e.g. exceeding 1022 FLOPs for text and/or image generation) as an initial threshold to assess whether a model qualifies as GPAI.
Important Distinction:
Although GPAI models are key components of AI systems, they are not AI systems themselves. A complete AI system also includes additional elements such as user interfaces or deployment mechanisms. A GPAI model becomes part of an AI system when it is integrated into an application or product that interacts with end users.