Docling vs dots.ocr
Pick Docling to ingest born-digital PDFs, Office files and HTML into structured JSON for RAG, since it handles many formats in one MIT-licensed library and runs on CPU. Pick dots.ocr to parse multilingual PDFs into Markdown with layout, tables as HTML and formulas as LaTeX, if you have a CUDA GPU. Docling's scan accuracy depends on the OCR engine you plug in, while dots.ocr is itself a 3B model with a published olmOCR-bench score of 79.1 (83.9 for dots.mocr).
Doclingibm/docling | dots.ocrrednote/dots-ocr | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 2.2 s / page | 1.5 s / page |
| Best for | Structured export for RAG | Multilingual document parsing |
| Inputs | PDF, images, DOCX | PDF, images |
| Languages | Latin scripts, CJK | 100+ languages |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | No | ✓ Yes |
| 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 Docling if
- You need one API for PDF, DOCX, PPTX, XLSX, HTML, EPUB, images, emails and XBRL.
- You want lossless JSON that keeps layout, reading order and table structure, with LangChain or LlamaIndex integrations.
- You have no GPU: CPU works on macOS, Linux and Windows, and a GPU only speeds up model stages.
Choose dots.ocr if
- You parse multilingual academic or business PDFs and want layout boxes, categories and text from a single model pass.
- You need tables as HTML and formulas as LaTeX in the output.
- You need charts or diagrams converted to SVG with the dots.mocr-svg variant.
Key differences
| Docling | dots.ocr | |
|---|---|---|
| What it is | Python library from IBM Research that converts documents into a DoclingDocument | 3B-parameter document parsing VLM (1.7B LLM backbone) from rednote hi lab |
| Licence | MIT (individual models have their own licences) | MIT, plus a separate dots.ocr license agreement file in the repository |
| How you run it | Local library on x86_64 and arm64; CPU works, GPU speeds up model stages | Single CUDA GPU with vLLM is the documented path; no VRAM figure stated |
| Input formats | PDF, DOCX, PPTX, XLSX, HTML, EPUB, images, emails, XBRL | Document images and PDFs (multilingual) |
| Output formats | Markdown, HTML, JSON, DocTags, plain text | JSON, Markdown, HTML, LaTeX, bounding boxes, SVG |
| Benchmarks | Not published | olmOCR-bench 79.1 (dots.ocr), 83.9 (dots.mocr), by the authors |
| Known failure | Version 2.143.1 drops words and numbers with full-page Tesseract OCR (issue #5254); some layouts merge neighbouring columns (issue #5132) | Runs of ellipses or underscores can make output repeat endlessly; complex tables and formulas stay difficult |
Sources
- Docling GitHub README
- Supported formats
- Issue #5254 OCR regression
- Issue #5132 reading order
- dots.ocr GitHub README
- dots.ocr blog / limitations
What each one is
dots.ocr
Single vision model from rednote that does layout detection and content recognition together.
dots.ocr 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": "ibm/docling","fallbacks": ["rednote/dots-ocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Docling alternativesdots.ocr alternativesBest OCR models in 2026