MinerU vs Florence-2
Pick MinerU to convert full PDFs and Office files into Markdown, HTML, LaTeX or JSON, especially Chinese and English academic papers. Pick Florence-2 when you only need plain text with line boxes from scene text or short English labels, or when you want a very small MIT-licensed model that also does captioning and detection. Florence-2 gives no document structure and handles Chinese badly, so it is not a page converter.
MinerUopendatalab/mineru | Florence-2microsoft/florence-2 | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 1 credit / page |
| Per 1,000 pages | $2.00 | $1.00 |
| Typical speed | 2.0 s / page | 0.8 s / page |
| Best for | Academic PDFs | Lightweight OCR with boxes |
| Inputs | PDF, images | images |
| Languages | 80+ languages | English |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | No |
| On OCRRouter | Available | Available |
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 takes PDF, images, DOCX, PPTX, XLSX, EPUB, OFD and HTML and renders Markdown, HTML, LaTeX, DOCX, EPUB or JSON.
- It has a CPU-only basic tier that runs on about 2 GB RAM, and a VLM tier that reaches 2.12 pages per second on one A100 with vLLM.
- It is listed as good for Chinese and English academic papers and mixed Office and PDF corpora for RAG, and scores 75.2 on olmOCR-Bench (2.5.4).
Choose Florence-2 if
- It is MIT licensed, while MinerU uses a custom licence and its VLM weights are tagged AGPL-3.0.
- At 0.23B (base) or 0.77B (large) parameters it is far smaller than the 1.2B MinerU2.5 VLM, which suits lightweight edge OCR in English.
- OCR_WITH_REGION returns a quadrilateral box for each text line, and the same model also handles captioning, detection and grounding.
Key differences
| MinerU | Florence-2 | |
|---|---|---|
| Licence | MinerU Open Source License (Apache-2.0 based with extra conditions); MinerU2.5 VLM weights tagged AGPL-3.0 | MIT |
| Model size | MinerU2.5 VLM: 1.2B | 0.23B (base), 0.77B (large) |
| Output | Markdown, HTML, LaTeX, DOCX, EPUB, JSON | Plain text and bounding boxes only; no Markdown, tables or LaTeX |
| Hardware | Basic tier: CPU with 2 GB RAM. VLM tiers: 16 GB RAM and a GPU with 8 GB+ VRAM | Not published |
| Chinese text | Listed as good for Chinese and English academic papers; converts full-width Chinese punctuation to half-width with no switch | Chinese output mostly wrong; the vocabulary lacks CJK characters |
| Handwriting | Not published | Spelling errors and missing spaces on cursive handwriting and French text in a Hugging Face blog test |
| Known failure | VLM tiers drop merged header rows in tables such as bank statement account lines; VLM pipeline can generate repeated content, including 300 KB figure blocks | Model card notes the continued-pretrained checkpoint used only 0.1B samples and might not be well trained |
Sources
- MinerU GitHub README
- olmOCR-Bench table
- Issue #5554 table header rows
- Florence-2-large model card
- HF discussion Chinese very bad
- HF blog OCR test
What each one is
MinerU
Open-source PDF to Markdown pipeline popular for academic and Chinese documents.
MinerU detailsFlorence-2
Small Microsoft vision model (0.23B and 0.77B) with OCR and OCR-with-regions tasks that return text with boxes.
Florence-2 detailsTry both on your own file
Run the file once with each model and compare the output.
Use one, fall back to the other
{"model": "opendatalab/mineru","fallbacks": ["microsoft/florence-2"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
MinerU alternativesFlorence-2 alternativesBest OCR models in 2026