Baidu open-sources Unlimited OCR 2 for parsing multi-page documents in one request
Article timeline
The articles in this story. Times are ET.
- PandailyNewsPickBaidu open-sources Unlimited OCR 2 for parsing documents of any length
AIBaidu has released and open-sourced Unlimited OCR 2, a document parsing model meant to read documents of any page count in one pass. Its code and weights are on GitHub and Hugging Face under the MIT license, and the model is also offered through Baidu AI Cloud. Baidu reports 84.6 on MPDocBench and 77.1 on MarathonDocBench, figures that are its own tests and have not been independently evaluated.
- AI EraNewsBaidu open-sources Unlimited OCR 2 document parsing model
AIBaidu has open-sourced Wenxin Unlimited OCR 2, a new-generation document parsing model, four months after its predecessor. The source text provided does not give benchmark scores, parameter counts, or pricing.
meng shaoXBaidu's Unlimited OCR 2 uses SR-SWA to parse long documents at fixed attention costAIBaidu has open-sourced Unlimited OCR 2, a 3B/A0.5B MoE document parsing model that targets documents of tens to hundreds of pages. The post explains its SR-SWA attention, which keeps text-to-visual interaction within a fixed 512-token reference set, and page-wise decoding that reuses KV tails. The author cites a claimed 7.0x speedup over Full Attention and 5.6x over R-SWA at around 100 pages.
ModelScopeOfficialUnlimited-OCR-2 parses entire multi-page documents in one requestAIPaddlePaddle's Unlimited-OCR-2 parses every page of a document in a single request, preserving cross-page context instead of treating pages as isolated images. The model reaches state-of-the-art results on MPDocBench and the hundred-page MarathonDocBench, and its Sparse Reference Sliding Window Attention keeps per-step decoding latency and memory nearly constant as documents grow. It supports native Transformers and SGLang serving under an MIT license.
- meng shaoXBaidu open-sources Unlimited OCR 2 for parsing long documents
AIBaidu has open-sourced Unlimited OCR 2, a 3B-parameter MoE model for parsing documents from dozens to hundreds of pages. The model uses SR-SWA to hold text-visual attention in a fixed-size set of reference tokens, which the author says keeps per-step decoding latency and memory nearly constant as documents grow. The author cites about 7.0x speedup over Full Attention and 5.6x over R-SWA at hundred-page scale.
- QbitAINewsPickBaidu open-sources Unlimited OCR 2, cutting peak KV cache use by about 96%
AIBaidu has released and open-sourced Unlimited OCR 2 on Hugging Face and GitHub. On a 100-page single-sequence test, peak KV cache use fell from 47GB under standard decoding to 1.77GB, a reduction of about 96%. The model adds Sparse Reference Sliding Window Attention (SR-SWA) and chunked decoding to process long documents continuously.
- LeiphoneNewsBaidu open-sources Unlimited OCR 2 for parsing documents of any length
AIBaidu released and open-sourced Unlimited OCR 2, which it says is the first OCR model to parse documents of any page count. It scored 84.6 on MPDocBench and 77.1 on MarathonDocBench, and its code and weights are available on GitHub and Hugging Face.
- MarkTechPostNewsBaidu open-sources Unlimited OCR 2 for hundred-page document parsing
AIBaidu has released Unlimited OCR 2, a 3.34B-parameter open-source OCR model under the MIT license that parses documents of hundreds of pages in one request. Its SR-SWA sparse attention caps attention at a fixed 512-token reference set, and Baidu reports 77.1 on MarathonDocBench against 33.7 for the next model. All benchmark figures are self-reported, and heading hierarchy is the weakest sub-score at 48.2.
SGLangOfficialBaidu releases open-source Unlimited OCR 2 for parsing long documentsAIBaidu has released Unlimited OCR 2 as an open-source model that parses documents of virtually any length within a finite context window. According to SGLang, its SR-SWA mechanism keeps decoding latency and memory nearly constant as documents grow, and the model reports state-of-the-art results on MPDocBench and MarathonDocBench. SGLang says it is the recommended serving path, with serving instructions in the model card and a cookbook to follow.
Baidu Inc.OfficialBaidu open-sources Unlimited OCR 2 for parsing documents of any lengthAIBaidu releases Unlimited OCR 2 as an open-source model that parses documents of virtually any length within a finite context window. Its SR-SWA mechanism condenses text-visual attention into a fixed-size set of working references, keeping per-step decoding latency and memory nearly constant as documents grow. Baidu says the model achieves state-of-the-art results on MPDocBench and MarathonDocBench, which cover documents from dozens to hundreds of pages.
