The note that requests aren't the same unit everywhere is the most useful line in the table. OCR.space caps pages per PDF, pdfRest counts calls and PDF.co counts credits, so the 200x spread shrinks or grows depending on the document you test with. Marking the unconfirmed RapidAPI figures as Not verified makes the whole comparison easier to trust.