Comment by jraedisch

First (Claude based) testing confirms not better or faster if one only needs text embeddings for retrieval or clustering. For music audio, MuQ-MuLan seems to remain the way to go? (Multimodal is ofc very useful for many other purposes.)