OCRRouter

MinerU vs TrOCR

Pick MinerU when the input is whole PDFs, scans or Office files and you need Markdown, HTML, LaTeX or JSON back. Pick TrOCR when you already have cropped single text-line images, mostly English handwriting or receipts, and want a recognizer you can fine-tune under an MIT licence. The main split is scope: MinerU parses documents, TrOCR reads one line at a time.

MinerUopendatalab/mineruTrOCRmicrosoft/trocr
TypeOpen sourceOpen source
Price2 credits / page1 credit / page
Per 1,000 pages$2.00$1.00
Typical speed2.0 s / page1.2 s / page
Best forAcademic PDFsHandwritten text lines
InputsPDF, imagesimages
Languages80+ languagesEnglish
Max pages per request500100
Tables✓ YesNo
HandwritingNo✓ Yes
Math✓ YesNo
Bounding boxes✓ YesNo
ConfidenceNoNo
FormsNoNo
Multilingual✓ YesNo
On OCRRouterAvailableAvailable

Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.

Which to choose

Choose MinerU if

  • You have PDFs, images, DOCX, PPTX, XLSX, EPUB, OFD or HTML and want one tool that converts them to Markdown, HTML, LaTeX, DOCX, EPUB or JSON.
  • You have no GPU: the basic ONNX tier runs on CPU with about 2 GB RAM, and the VLM tier reaches 2.12 pages per second on one A100 with vLLM.
  • Your corpus is Chinese and English academic papers or mixed Office and PDF files for RAG.

Choose TrOCR if

  • You need to read English handwritten lines: the large model reports 2.89 character error rate on the IAM handwriting set, per the paper.
  • You want to fine-tune a recognizer on your own line images, starting from separate checkpoints for handwritten (IAM), printed (SROIE) and scene text.
  • You need a plain MIT licence and a few lines of Hugging Face transformers code.

Key differences

MinerUTrOCR
InputPDFs, images, DOCX, PPTX, XLSX, EPUB, OFD and HTMLSingle text-line images only, so you need a separate line detector
OutputMarkdown, HTML, LaTeX, DOCX, EPUB and JSONPlain text, with no layout, table or page-level output
LicenceMinerU Open Source License (Apache-2.0 based with extra conditions); MinerU2.5 VLM weights are tagged AGPL-3.0MIT (microsoft/unilm repository)
Model size and hardwareMinerU2.5 VLM is 1.2B parameters; basic tier runs on CPU with 2 GB RAM; VLM tiers need 16 GB RAM and a GPU with 8 GB+ VRAMBase 334M, large 558M parameters; slow versus CRNN recognizers on CPU
Benchmark75.2 on olmOCR-Bench overall (MinerU 2.5.4); 45.9 ParseBench mean (MinerU2.5)2.89 character error rate on IAM handwriting (TrOCR-large), reported by the authors
LanguagesGood for Chinese and English papers; converts full-width Chinese punctuation to half-width with no switchReleased checkpoints are English-focused; other languages need fine-tuning
Known failureVLM tiers drop merged header rows in tables such as bank statement account lines; the VLM pipeline can generate repeated content, including 300 KB figure blocksCannot read full pages or tables at all, since it only takes one text line

Sources

What each one is

MinerU

Open-source PDF to Markdown pipeline popular for academic and Chinese documents.

MinerU details

TrOCR

Transformer text recognition model. Reads single text lines, including handwriting.

TrOCR details

Try both on your own file

Run the file once with each model and compare the output.

PDF, PNG, JPG, WebP, HEIC. Up to 20 MB each, 10 files.No file?

Use one, fall back to the other

{"model": "opendatalab/mineru","fallbacks": ["microsoft/trocr"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

MinerU alternativesTrOCR alternativesBest OCR models in 2026