Multimodal decision models and very expensive piles of PDFs Antfly engineer Rowan Copley published a cost comparison showing that extracting text from 3.5 million pages of PDFs ranges from roughly $3,000 to $196,875 depending on the service and mode, with AWS Textract's Forms + Tables mode at $190,000 and LlamaParse's Agentic Plus at $196,875 as the most expensive options and Textract text OCR, Azure Document Intelligence Read and Google Document AI Enterprise OCR near $3,000–$5,250. Copley, who is scaling Antfly's multimodal ingestion, said the company needs to run the work locally and argued most pages need only cheap heuristic checks rather than OCR, since a modern OCR model such as LightOnOCR-3-4B misread "8 1%" as 81% and "19 4%" as 194% on a 1989 backpropagation paper scan. Routing pages through a larger model like GPT-6-Luna transcribed that page correctly for $0.38, which Copley estimated would put the full workload at about $13,000. Back to Rowan Copley https://antfly.io/team/rowan · Lab Multimodal decision models and very expensive piles of PDFs What would it take to index a massive corpus of PDFs? Like, for example, the entirety of the Epstein files? This is a question I found myself asking recently as we're scaling Antfly's ability to ingest multimodal data. Like, here's a table I had Claude make that estimates the total cost of extracting the Epstein files' text on various services: | Service / mode | Published USD per 1,000 pages | Cost for 3.5M pages