Baidu open-sources Unlimited-OCR-2 for long-document parsing
First merged summary.
Baidu has published Unlimited-OCR-2, an OCR model aimed at parsing documents of unbounded length within a finite context window.
Baidu published Unlimited-OCR-2, an OCR model that uses Sparse Reference Sliding Window Attention to parse documents of unbounded length inside a finite context window. The project says per-step latency and memory stay nearly constant as documents grow, and claims state-of-the-art results on MPDocBench and MarathonDocBench. Weights are on Hugging Face and ModelScope, with Baidu Cloud availability, Transformers and SGLang inference, and a Hugging Face demo. A later MarkTechPost item framed the release as open source and aimed at hundred-page parsing, but its captured page contained only a cookie notice and no further specs.
- SR-SWA keeps a fixed working set, such as 512 tokens.
- MarathonDocBench, a hundred-page parsing benchmark, was released the same day.
- Multi-page inference parses every page in one request.
- Single-page mode uses dynamic tiling, called gundam mode, by default.
- The model card lists Transformers 4.52.1 and a custom SGLang wheel.
Timeline
dated events in the reporting · oldest first- Oct 26, 2026Baidu publishes Unlimited-OCR-2 and MarathonDocBench. · Hugging Face trending models
What each outlet adds
beyond the shared facts- Hugging Face trending modelsArchitecture, benchmarks, serving stack, and inference modes.
- Vendors
- Baidu
- Organizations
- ModelScope
- AI models
- Unlimited-OCR-2Unlimited OCR 2
Coverage
oldest first · one row per article · UTC- · 4d ago
Hugging Face trending modelsFirst report
Baidu releases Unlimited-OCR-2 for unbounded document parsing
SR-SWA keeps a fixed working set, such as 512 tokens, and the release adds the MarathonDocBench benchmark.
- · 3d ago
MarkTechPost
Baidu open-sources Unlimited OCR 2 for long documents
The headline says the release targets hundred-page document parsing; no further specs were in the text.