LlamaParse looked great on paper. $10 for 1k pages a month? Fine.
Then reality hit:
* Their "page" is a sliding scale based on complexity. A dense PDF with tables? That's 3-4 "pages."
* Processing 100 actual documents blew through my monthly quota in a week.
* API errors still count against your usage. Found that out the hard way.
Now the bill is creeping toward "enterprise" territory.
What are people using instead? My needs:
* Parsing mixed docs (PDFs, PPT, Word) for RAG.
* Must have a predictable, actual per-document cost.
* Open-source self-hosted is fine, if the setup isn't a full-time job.
Unparser? Nougat? DIY with some combo of libraries? What's the real TCO?
always ask for a multi-year discount
The "sliding scale" pricing for complex pages is the real trap. It turns cost predictability into a guessing game.
For a stable per-document cost, self-hosting is the only reliable path. Unstructured.io's open-source library is the core of many commercial parsers. You can run it yourself via their API containers, and your cost becomes compute time, which is finite and known. It handles your listed formats.
The setup isn't trivial, but it's not a full-time job. The TCO shifts from opaque API charges to your DevOps overhead and cloud compute. For consistent, high-volume parsing, that trade-off usually wins.