Performance & Cost
For document parsing, performance and cost are tightly coupled: your default mode, how many pages you process, and whether you route by document type.
Key facts
Section titled “Key facts”- PDFs are priced per page (1 credit per page)
fast/auto/ocrchanges latency and success ratemaxPagesis the simplest cost/time cap
Recommended: learn your distribution first
Section titled “Recommended: learn your distribution first”- sample ~20 PDFs
- estimate how many are text-native vs scanned vs table-heavy
- pick defaults per class instead of one-size-fits-all
Recommended: two-stage parsing
Section titled “Recommended: two-stage parsing”- stage 1:
auto + maxPages: 5for quick probing - stage 2: for scanned/complex docs, switch to
ocror increasemaxPages
Common pitfalls
Section titled “Common pitfalls”- defaulting everything to
ocrwithout sampling - not keeping failure examples for regression testing
Related: