We’ve Added Mistral OCR 4

We have finished adding Mistral OCR 4 to the site. It now appears in the model picker next to OCR 3, so there is nothing new to learn: choose the model, upload your PDF or image, and run it as usual.

So, what is actually different?

OCR 3 is very good at turning a document into clean text and Markdown. OCR 4 goes further with the structure of the page. It can return bounding boxes, label blocks such as titles and tables, and provide confidence information. In plain terms, it gives software more clues about where something was on the original page—not only what it said.

That extra information matters for jobs such as linking an answer back to a spot on the page, finding text that needs manual review, preserving table context, or preparing documents for RAG. If you only need readable text, you may not need it.

OCR 3 is not going away

Newer does not automatically mean better for every upload. OCR 3 still costs less here and remains a sensible choice for ordinary text, Markdown, and table extraction. It uses 1 credit per processed page; OCR 4 uses 2.

Our rule of thumb is simple: start with OCR 3 when you mainly want the text. Pick OCR 4 when the document layout or confidence data will be useful after extraction. You can switch models before every upload, so you do not have to commit to one.