Baidu Ships 'Unlimited-OCR' for One-Shot Long-Horizon Document Parsing — 343 Points on HN

GitHub (Baidu) / Hacker News·medium signal

Baidu's open-source Unlimited-OCR landed on the HN front page with ~343 points / 84 comments, pitched as one-shot parsing of long, multi-page documents rather than the page-at-a-time chunking most OCR pipelines rely on. For builders running RAG or document-intelligence stacks, long-horizon OCR that holds context across a whole document is a meaningful ingestion-quality upgrade. Released to GitHub; worth benchmarking against existing layout-aware extractors.