You can get the best of both worlds with a small embedding model
Show me an embedding model that runs as fast as an optimized regex engine and I'll buy you a beer.
Show me an embedding model that runs as fast as an optimized regex engine and I'll buy you a beer.