Reportedly: Baidu open-sources Wenxin Unlimited OCR 2 document parsing model with 'unlimited' page support
Reported by IT Home · not yet confirmed by the company or a second independent outlet. We update this page when it is.
Baidu released and open-sourced Wenxin Unlimited OCR 2 on October 27, 2026, extending document parsing beyond dozens of pages to 'unlimited' and parsing over 3,600 pages of the Harry Potter series in one pass. It scored 84.6 on MPDocBench and 77.1 on MarathonDocBench, both end-to-end SOTA, IT Home reported.
China context
- Original name
- 文心 Unlimited OCR 2
- Outside China
- Open weights · ithome.com
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers can integrate the open-weights model for high-volume document parsing without per-page API costs, but should verify performance on their own document types.
- For investors
- Baidu's open-source release intensifies competition in document AI, potentially pressuring commercial OCR API pricing and market share.
Translated from Chinese. Quotes and facts link to the original sources.
Baidu officially released and open-sourced the Wenxin Unlimited OCR 2 model on October 27, 2026. The model pushes document parsing limits from dozens of pages to 'unlimited', capable of parsing over 3,600 pages of the complete Harry Potter series in a single pass. It natively supports layout element segmentation and recognition, cross-page text merging, and table recognition and merging, delivering more processing with the same compute, faster and cheaper. Baidu reported scores of 84.6 on MPDocBench and 77.1 on MarathonDocBench, both achieving end-to-end state-of-the-art performance.
The model's 'unlimited' page handling likely relies on efficient long-context processing and memory management, enabling cross-page text and table merging without truncation. The reported SOTA scores on MPDocBench and MarathonDocBench suggest strong performance on multi-page and hundred-page document parsing tasks.
Developers and enterprises outside China gain a new open-weights option for large-scale document parsing, potentially reducing reliance on proprietary OCR APIs and lowering per-page processing costs. Competitors in document AI must now match or exceed this capability to retain customers with high-volume document workflows.
Open-sourcing the model allows businesses to self-host document parsing at scale, avoiding per-page API fees and data egress concerns. The ability to process thousands of pages in one pass can streamline workflows in legal, financial, and publishing sectors.
A verifiable next signal is whether independent benchmarks or user reports confirm the claimed SOTA scores and 'unlimited' page handling on real-world documents. Another signal is adoption by major document processing platforms or cloud services outside China.