Glossary · AI Act
GPAI (General-Purpose AI)
AI models trained for broad capability and usable across many tasks, subject to their own obligations under the AI Act since August 2025.
Explained in depth: AI Act overview
A general-purpose AI model displays significant generality and competently performs a wide range of distinct tasks, regardless of how it is placed on the market; large language models are the canonical example. Since 2 August 2025, GPAI providers must maintain technical documentation, provide information to downstream providers who build on the model, put in place a policy to comply with EU copyright law, and publish a summary of the content used for training. Models designated as posing systemic risk, based on their capabilities or the compute used in training (with a threshold of 10^25 floating-point operations creating a presumption), face additional duties: model evaluations, adversarial testing, systemic risk assessment and mitigation, serious incident reporting, and cybersecurity protection. A voluntary code of practice helps providers demonstrate compliance.
Why it matters
Most companies consume GPAI rather than build it, but the rules still matter downstream: documentation from your model provider is what makes your own AI Act compliance possible, and building products on top of a model can, in some circumstances, make you a provider yourself.