Whilst BERT is state of the art, I think Word2Vec is pretty good too. I think it is still relevant in following cases:

Not everyone has to implement a state of the art model for their business problem, Word2Vec can be trained on custom data too and it gives pretty good results. The trade-off between accuracy, model understanding and time taken all comes down to business needs.

Another reason behind the relevance is the academics. You always have to understand the basics, before going to an advanced level.