AFAIK, the “large” qualifier came when transformers allowed to scale the size of language models compared to the recurrent models that where in fashion before.

AFAIK, the “large” qualifier came when transformers allowed to scale the size of language models compared to the recurrent models that where in fashion before. And although BERT isn't large by today's standard, it was large enough for the time.