FreeLLMBaidu launches Unlimited-OCR-2 on Hugging Face with unlimited‑horizon parsing
The new OCR model uses Sparse Reference Sliding Window Attention to handle ultra‑long documents and is available on Hugging Face, ModelScope and Baidu Cloud.
Baidu released Unlimited-OCR-2 on Hugging Face on 28 October 2026. The model page shows 109,519 downloads and 496 likes.
Unlimited-OCR-2 introduces Sparse Reference Sliding Window Attention (SR‑SWA), which limits attention to a fixed set of 512 reference tokens regardless of document length.
The model claims state‑of‑the‑art results on long‑document benchmarks such as MPDocBench and the newly released MarathonDocBench, which targets hundred‑page parsing.
It is distributed under an MIT license and can be run with Transformers or the recommended SGLang serving stack.
Baidu made the model available on ModelScope, Baidu Cloud and a Hugging Face Space demo created by AK.
What it means for you
Unlimited-OCR-2 lets developers parse multi‑page documents in a single request, reducing latency and memory use for very long texts.
Sources
- Baidu releases Unlimited-OCR-2 on Hugging Face, Hugging Face trending (official), 26 Oct 2026
Written automatically from the sources above, then checked by our code: every number and name must appear in them, and no sentence may copy a source. Spotted a mistake? Report it.
Which AI can you use for free today?
Every free AI model, ranked by blind human votes, with the memory and speed it needs on your own GPU.