OCRRouter

MinerU vs SmolDocling

Pick MinerU for Chinese and English academic papers and mixed PDF and Office corpora, where one tool reads PDF, images, DOCX, PPTX, XLSX, EPUB, OFD and HTML. Pick SmolDocling only for small English experiments, because it is a 256M preview model and its authors point users to Granite-Docling. MinerU's main cost is a custom licence and AGPL-tagged VLM weights. SmolDocling's main cost is that it is English only and can loop on some images.

MinerUopendatalab/mineruSmolDoclinghuggingface/smoldocling
TypeOpen sourceOpen source
Price2 credits / page1 credit / page
Per 1,000 pages$2.00$1.00
Typical speed2.0 s / page1.0 s / page
Best forAcademic PDFsVery cheap page conversion
InputsPDF, imagesPDF, images
Languages80+ languagesMainly English
Max pages per request500500
Tables✓ Yes✓ Yes
HandwritingNoNo
Math✓ Yes✓ Yes
Bounding boxes✓ Yes✓ Yes
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

  • It accepts PDF, images, DOCX, PPTX, XLSX, EPUB, OFD and HTML in one tool, so a mixed corpus needs no separate converters.
  • Its basic ONNX tier runs on CPU with about 2 GB RAM, so you can start batch conversion without a GPU.
  • It is a fit for Chinese and English academic papers, and it scores 75.2 overall on olmOCR-Bench (MinerU 2.5.4).

Choose SmolDocling if

  • At 256M parameters it runs on modest hardware, and it takes about 0.35 seconds per page on an A100 with vLLM.
  • It is licensed Apache-2.0 per its model summary (the metadata tag says CDLA-Permissive-2.0), which is simpler than MinerU's custom licence and AGPL-tagged weights.
  • One model handles text, tables (OTSL), code, formulas and charts, and it returns location tags and bounding boxes.

Key differences

MinerUSmolDocling
LicenceMinerU Open Source License (Apache-2.0 based with extra conditions); the MinerU2.5 VLM weights are tagged AGPL-3.0Apache-2.0 per model summary (metadata tag says CDLA-Permissive-2.0)
Model size and hardwareVLM: 1.2B parameters. Basic tier runs on CPU with 2 GB RAM; VLM tiers need 16 GB RAM and a GPU with 8 GB+ VRAM256M parameters, small enough for modest hardware
Input formatsPDF, images, DOCX, PPTX, XLSX, EPUB, OFD and HTMLPage images converted by the VLM; other input types Not published
Output formatsMarkdown, HTML, LaTeX, DOCX, EPUB, JSONDocTags, Markdown, HTML, JSON, LaTeX, bounding boxes
LanguagesChinese and English academic papers are a stated fitEnglish only
SpeedVLM tier reaches 2.12 pages per second on one A100 with vLLMAbout 0.35 seconds per page on an A100 with vLLM
Known failureVLM pipeline can repeat content (300 KB figure blocks, issue #5728) and drops merged header rows in tables such as bank statements (issue #5554)Endless token repetition and missing closing tags are typical defects; output loops on some images, making generation very slow (Docling issue #1515)

Sources

What each one is

MinerU

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

MinerU details

SmolDocling

Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.

SmolDocling 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": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

MinerU alternativesSmolDocling alternativesBest OCR models in 2026