OCR models now digitize a million pages for under $130 as costs drop fivefold in 18 months
Reported by The Decoder1 source ↓
Pulling text out of scanned books and records has gotten a lot cheaper with freely available AI models. The article OCR models now digitize a million pages for under $130 as costs drop fivefold in 18 months appeared first on The Decoder .
Key points
- The newest model tested cost just under 10 cents per 1,000 pages, compared with 51 cents for the oldest
- Estimated GPU rental for a million pages ranged from $65 to $130, depending on the source material
- The newest model read book pages at least as accurately as the oldest, but some models produced text on blank pages or repeated content
- Running models more efficiently saved about as much as switching to newer models, van Strien said
Details
The Decoder reports that Daniel van Strien of Hugging Face compared eight OCR models using 850 pages from old natural history books, the 1771 Encyclopaedia Britannica, and typed and printed archival documents. He ran them on the same rented Nvidia RTX PRO 6000 GPU, at $2.75 per hour. The oldest model, RolmOCR, dates to April 2025; the newest, LightOnOCR-3-0.8B, dates to October 2026.
Van Strien said operating changes—including using smaller scans, distributing pages more evenly, updating software, and using a faster GPU—saved about as much as changing models, or more. The newest model matched or exceeded the oldest model’s accuracy on book pages, but some models hallucinated text on blank pages or repeated content. The comparison found that the five models between the oldest and newest had similar costs.
Written from coverage as of
Images
- The Decoder
- The Decoder
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh