What Are GPAI Models?
General-Purpose AI (GPAI) models are AI models trained on broad data at scale that can be adapted to a wide range of tasks. This includes large language models (LLMs) like GPT-4, Claude, Gemini, and open-source models like Llama and Mistral. The AI Act introduced specific obligations for GPAI model providers starting August 2, 2025.
Standard GPAI Obligations
- Prepare and maintain up-to-date technical documentation
- Provide information and documentation to downstream providers integrating the model
- Establish a copyright compliance policy, particularly regarding the Text and Data Mining opt-out
- Publish a sufficiently detailed summary of the training data content
Systemic Risk GPAI Models
GPAI models that pose systemic risks (defined as those with high-impact capabilities, or models trained with more than 10^25 FLOPs of compute) face additional obligations:
- Perform model evaluations including adversarial testing
- Assess and mitigate possible systemic risks
- Track, document, and report serious incidents
- Ensure adequate cybersecurity protections
GPAI Code of Practice
The EU AI Office has published a voluntary GPAI Code of Practice that provides detailed guidance on meeting these obligations. Adhering to the Code provides a presumption of conformity with the AI Act's GPAI requirements. Major model providers have already become signatories.