This article is the first in a series exploring the obligations of providers of General-Purpose AI (GPAI) models under the EU Artificial Intelligence Act (AI Act). In this part, I focus on the definition of GPAI models and the application of the rules. Future articles will examine the obligations of providers, the grandfathering rules, and other key aspects of the AI Act relevant to GPAI model providers.

The AI Act, adopted in 2024, introduces a comprehensive regulatory framework for the development, placing on the market and use of AI systems in the European Union. Among its key innovations is the regulation of providers of GPAI models. GPAI models are large-scale models that can be adapted for a wide range of downstream use cases.

The European Commission has issued the “Guidelines on the scope of the obligations for general-purpose AI models” (Guidelines) to clarify the application of these obligations for the industry. These Guidelines are not binding. Nevertheless, the European Commission notes that the Guidelines reflect its interpretation and application of the AI Act, and that it will rely on the Guidelines in its enforcement actions against providers of GPAI models.

Below, I address the most common questions you may have regarding GPAI models:

1. On what date do the provisions of the AI Act applicable to providers of GPAI models come into force?

The provisions entered into force on 2 August 2025.

2. How can I determine whether my GPAI model falls under the scope of the AI Act?

Article 3(63) of the AI Act defines a GPAI model as an AI model that demonstrates “significant generality” and is “capable of competently performing a wide range of distinct tasks.”. To provide further practical clarity, Section 2.1 of the Guidelines introduces indicative criteria. A model is considered a GPAI model if:

• its training compute exceeds 10²³ FLOPs; and
• it is capable of generating language (in text or audio form), text-to-image, or text-to-video.

However, if a model meets the above criterion but, exceptionally, does not display significant generality or is not capable of competently performing a wide range of distinct tasks, it will not be considered a GPAI model. Conversely, if a model does not meet the criterion but, exceptionally, demonstrates significant generality and can competently perform a wide range of distinct tasks, it will still be regarded as a GPAI model.

The Guidelines include examples of models that fall out of scope. Here are two of them, but you can check the Guidelines for the full list:

• A model is trained specifically to transcribe speech into text, using 10²⁴ FLOPs. Since it can generate text and its training compute exceeds 10²³ FLOPs, the criterion provided in the Guideline suggests that it should qualify as a GPAI model. However, because the model can competently perform only a narrow task (speech transcription), it is not in fact a GPAI model.
• A model is trained specifically to play chess or video games, using 10²⁴ FLOPs. Although its training compute exceeds 10²³ FLOPs, if the model cannot generate text, speech, text-to-image, or text-to-video, the criterion provided in the Guidelines indicates that it should not be considered a GPAI model. Moreover, since the model can competently perform only a narrow set of tasks (chess or video games), this confirms that it is not a GPAI model.

3. What is a GPAI model with systemic risk?

GPAI models with systemic risk form a special class of GPAI models under the AI Act.

Under Article 51(1) of the AI Act, a GPAI model is classified as a GPAI model with systemic risk if it meets either of the following conditions:

• High-impact capabilities – the model has capabilities that match or exceed those of the most advanced models (Article 3(64) AI Act); or
• Commission decision – the European Commission, either on its own initiative or following a qualified alert from the scientific panel, determines that the model has capabilities or an impact equivalent to those described above, taking into account the criteria set out in Annex XIII of the AI Act.

From the moment when a GPAI model meets either of the two conditions above, the model is classified as a GPAI model with systemic risk and its provider must comply with the relevant obligations of the AI Act.

4. At what point do the rules of the AI Act apply to me as a GPAI model provider?

For the AI Act to apply to a provider of a GPAI model, the model must be placed on the market, that is, made available for the first time on the market of the European Union (EU). Therefore, before placing your GPAI model on the EU market, you must ensure compliance with the AI Act.