Hugging Face and EleutherAI created the FineBooks BHL OCR Leaderboard to test whether open-source OCR models can accurately digitize historical books. The project benchmarked 14 open-weight models against 2,165 ground-truth pages from the Biodiversity Heritage Library, transcribed to an error rate of roughly one in two thousand characters by human experts, using Character Error Rate as the primary metric. The leading model reached 97.6% accuracy while costing under one dollar per thousand pages to process, suggesting that large-scale re-digitization of historical archives is now economically viable.
