[qeternity]

I'm sure this will change over time, but I have yet to see an LMM that performs (on average) as well as decent text extraction pipelines.

Text embeddings for text also have much better recall in my tests.