1littlecoder · 2026-10-27 · notable
1littlecoder — 'Unlimited OCR 2 is here!'
1littlecoder covers Unlimited OCR 2, Baidu's new MIT-licensed 3B document-parsing model that uses Sparse Reference Sliding Window Attention to read whole long documents in one request.
1littlecoder looks at Unlimited OCR 2, Baidu's open OCR model that parses a whole book in one request.
What is it?
'Unlimited OCR 2 is here!' went up on the 1littlecoder channel on 27 October 2026, the same day Baidu published Unlimited OCR 2 on Hugging Face. The model is a 3-billion-parameter document-parsing vision-language model released under the MIT license.
How does it work?
The new piece in version 2 is Sparse Reference Sliding Window Attention (SR-SWA). Per Baidu's model card, SR-SWA squeezes attention from the generated text to the page images into a fixed-size set of working references, so per-step decoding latency and memory stay nearly constant however long the document is.
Why does it matter?
Most OCR pipelines split a PDF into pages and lose cross-page context. Baidu claims state-of-the-art results on its multi-page MPDocBench and hundred-page MarathonDocBench benchmarks, which makes a small open model a candidate for parsing full reports and books locally.
Who is it for?
developers building document-parsing and RAG pipelines
Try it
baidu/Unlimited-OCR-2 on Hugging Face