mishal brands

Unlimited OCR 2: developer tools for developers and technical teams

Visit official site

Official site link, not an affiliate link. We do not have an approved referral link for Unlimited OCR 2.

Unlimited OCR 2 helps you make documents and slides by this route: baidu’s free, MIT-licensed model that turns very long documents into text in one run. It is best for developers and technical teams who need to turn long PDFs, reports or scanned books into text and tables. People also use it to build software. A free plan is available.

Screenshot of Unlimited OCR 2 homepage
Homepage

What it does

Most OCR models handle a page or a short document at a time. Baidu says Unlimited OCR 2 uses a new attention method, Sparse Reference Sliding Window Attention, that keeps only a fixed set of working references, so it can keep reading without its memory use growing with the document. The model is about 3.3 billion parameters on Hugging Face. Baidu also released MarathonDocBench, a benchmark for very long document parsing, and the model is on ModelScope and Baidu Cloud, per its model card.

Official site:huggingface.co/baidu/Unlimited-OCR-2.

Editorial note

Baidu released Unlimited OCR 2 on October 27, 2026. It is an open, MIT-licensed vision-language model that Baidu says can parse documents of unlimited length in a single run, keeping per-step memory and speed nearly constant however long the document is.

Who it’s for

Key use cases

What you can do with it

Screenshot of Unlimited OCR 2 website
Official page
Screenshot of Unlimited OCR 2 website
Official page

How to use Unlimited OCR 2: step-by-step tutorial

Five to eight steps from the account to a first useful result. Where the official guide names a control, the step uses that name. Where it does not, the step stays general.

  1. Open the free Unlimited OCR 2 demo on Hugging Face Spaces and upload a sample document.
  2. Check that tables and reading order come out the way you need.
  3. For real use, follow the setup on the Hugging Face model card or the GitHub repo.
  4. Run it on a machine with a GPU, using Transformers or vLLM.
  5. Spot-check the text against the original before you rely on it.

These steps follow the official guide: Unlimited OCR 2 model card (Hugging Face).

Frequently asked questions

What is Unlimited OCR 2?

An open OCR model from Baidu, released October 27, 2026, built to turn very long documents into text in a single run, per its model card.

Is Unlimited OCR 2 free?

Yes. It is published under the MIT license on Hugging Face, checked October 27, 2026. You pay only for the hardware you run it on.

Do I need a GPU?

For practical speed, yes. It is a model of about 3.3 billion parameters that you run yourself. The Hugging Face demo lets you try it without one.

What makes it different?

Baidu says its Sparse Reference Sliding Window Attention keeps decoding speed and memory nearly constant regardless of document length.

Where can I get it?

On Hugging Face (baidu/Unlimited-OCR-2), with code on GitHub (baidu/Unlimited-OCR), and on ModelScope and Baidu Cloud, per Baidu.

Pricing

Prices checked on the official pricing page.

Open model (MIT)

Free

Download the weights from Hugging Face and run them yourself with Transformers or vLLM

Web demo

Free

Try it in the browser on Hugging Face Spaces

Facts

Access
Ongoing free plan
Paid plans from
Free
Checked
Category
Category: Development tools (31 in the library).
On the record
Baidu publishes Unlimited OCR 2 on Hugging Face under the MIT license, so the model weights are free to download and use. You run it on your own hardware, and there is a free web demo on Hugging Face Spaces.
Sources