Unlimited OCR

Unlimited OCR is a model from Baidu that performs long-horizon document parsing in a single pass, eliminating the need to split documents into chunks. It achieves a score of 456 points on a relevant benchmark, indicating strong performance on complex, lengthy documents. The model addresses a key pain point in OCR workflows where traditional systems struggle with context loss across page or section boundaries. By processing entire documents at once, it preserves layout and semantic coherence, making it suitable for tasks like invoice extraction, legal document analysis, and archival digitization. The evidence comes from a GitHub repository and community discussion on Hacker News, suggesting a fresh launch with high commercial intent. While the model appears to be open-source, specific usage details are still emerging.

Spotted on Hacker News via a GitHub repository from Baidu, indicating a fresh open-source launch with strong community interest in long-document OCR.