Baidu Unlimited OCR 2: open 3B model that parses hundred-page documents in one request
Baidu released Unlimited OCR 2 on 27 Oct 2026: a 3.34B-parameter, MIT-licensed OCR model that reads documents spanning hundreds of pages in a single request. It uses Sparse Reference Sliding Window Attention (SR-SWA), which keeps a fixed 512-token reference set so decoding time and memory stay nearly constant as documents grow; Baidu reports 77.1 on MarathonDocBench.
Why it matters
Long-document OCR without the memory blow-up: a fixed reference pool replaces full attention over the document, so a whole book or filing can be parsed in one pass on modest hardware.
What you could build with it
A legal-tech team could build a contract-digitisation service that ingests hundred-page scanned agreements in one request and returns structured clauses, tables and cross-references, without splitting files into pages first.
Does it hold up?
Too early to judge: released today and all benchmark numbers are Baidu's own, but the MIT licence and published Transformers/SGLang run path mean independent teams can verify it within days.
Built with Baidu Unlimited OCR 2
- Baidu open-sources Unlimited OCR 2 for hundred-page document parsingarticle · Launch coverage with benchmark detail: 77.1 on MarathonDocBench against 33.7 for the next model, and a warning that heading hierarchy is the weak spot.
- Baidu releases open-source OCR model for very long documentsarticle · Independent writeup confirming the MIT licence, Transformers/SGLang run options, and that Baidu's benchmark claims are vendor-reported, not independently measured.
Learn more
First spotted on hf: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73NVIDIA Isaac Lab 3.0 GA + Isaac OS open runtime + Agile ONE SN humanoidMajor Isaac platform expansion at GTC Berlin Oct 21: Isaac Lab 3.0 is generally available as the open-source robot…models · JEV 0.73StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73NVIDIA Nemotron 3.5 Retriever collection: Embed 8B/1B, Rerank 8B, Parse 631MNVIDIA's Oct 21 collection drop for agentic retrieval: Nemotron-3.5-Embed in 8B BF16, 1B BF16 and 1B NVFP4…models · JEV 0.72Kimi K3.1 Pro: Moonshot's 2.8T open-weight flagship officially launchesMoonshot AI officially launched Kimi K3.1 Pro on Oct 23, a mixture-of-experts model with 2.8 trillion total / 104…models · JEV 0.72
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.