For determining that a general-purpose AI model has capabilities or impact equivalent to Article 51(1)(a), the Commission shall consider:
(a) number of parameters of the model
(b) quality or size of the data set (e.g., measured through tokens)
(c) amount of computation used for training (floating point operations, estimated cost, time, energy consumption)
(d) input and output modalities (text to text, text to image, multi-modality) and state of the art thresholds
(e) benchmarks and evaluations of capabilities:
- number of tasks without additional training
- adaptability to learn new tasks
- level of autonomy and scalability
- tools accessible to it
(f) high impact on internal market (presumed if available to at least 10,000 registered business users in the Union)
(g) number of registered end-users