Baidu open-sources Unlimited OCR 2, a document parser with near-constant decoding cost

Baidu announced Unlimited OCR 2 on X on October 26, with code on GitHub and weights on Hugging Face. The model is built to parse documents of virtually any length within a finite context window.

Baidu says its new SR-SWA attention mechanism condenses text-visual attention interactions into a fixed-size set of working references, so per-step decoding latency and memory footprint stay nearly constant as documents grow longer. A relayed summary of the release says the model has 3.34B parameters, is released under the MIT license, and uses a fixed working set of 512 tokens; those details come from a user's post, not Baidu's announcement.

Claims from the announcement, not independently measured:

• State-of-the-art results on MPDocBench and MarathonDocBench, which cover documents from dozens to hundreds of pages.