Tuesday, July 21, 2026

Unlimited-OCR

★21,466

One-shot long document parsing that pushes past short-page OCR demos.

Unlimited-OCR

Why it matters

Document AI still breaks when the page count and layout complexity go up. Unlimited-OCR from Baidu is an open long-horizon parser aimed beyond short single-shot OCR: multi-page and PDF paths, long generation budgets, vLLM and SGLang serving, and a public model on Hugging Face. The practical win is open weights and inference for real multi-page parsing work, not another screenshot-to-text toy.

ocrdocumentsvision
Source ↗

Via github

X