Docling vs MiniCPM-V 4.5
Pick Docling to ingest born-digital PDFs and Office files into a RAG pipeline that needs structured JSON with layout. Pick MiniCPM-V 4.5 to read scans and images with a single local model, on a single GPU or on-device with quantized builds. The main split is that Docling is a conversion library that needs an OCR engine for scans, while MiniCPM-V is a vision model that reads the page itself but gives no coordinates.
Doclingibm/docling | MiniCPM-V 4.5openbmb/minicpm-v | |
|---|---|---|
| Type | Open source | Vision model |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 2.2 s / page | 2.5 s / page |
| Best for | Structured export for RAG | OCR on edge devices |
| Inputs | PDF, images, DOCX | images |
| Languages | Latin scripts, CJK | 30+ languages |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | ✓ Yes |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | No |
| 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 need lossless JSON that keeps layout, reading order and table structure for downstream tools.
- You want LangChain or LlamaIndex integrations and an MIT-licensed library that is a common RAG ingestion default.
Choose MiniCPM-V 4.5 if
- You want a single model that does OCR and document Q&A locally, without choosing and tuning a separate OCR engine.
- You need on-device or edge OCR, since int4, GGUF and AWQ builds run on llama.cpp and vLLM.
- You want Apache-2.0 weights and code, with OpenBMB claiming leading OCRBench results and state of the art PDF parsing on OmniDocBench.
Key differences
| Docling | MiniCPM-V 4.5 | |
|---|---|---|
| What it is | Python library from IBM Research that converts documents into a structured DoclingDocument | 8.7B vision language model from OpenBMB, built on Qwen3-8B and SigLIP2-400M |
| Licence | MIT (individual models have their own licences) | Apache-2.0 for weights and code |
| Input formats | PDF, DOCX, PPTX, XLSX, HTML, EPUB, images, emails, XBRL | Images and documents passed to a VLM; no format list published |
| Output formats | Markdown, HTML, JSON, DocTags, plain text | Plain text, Markdown, JSON |
| Layout and coordinates | Lossless JSON keeps layout, reading order and table structure | No bounding box or layout output |
| Scanned pages | Accuracy depends on the OCR engine you plug in (EasyOCR, Tesseract, RapidOCR) | The model reads the page itself; benchmark scores are published only as images |
| Known failure | Version 2.143.1 drops words and numbers with full-page Tesseract OCR (issue #5254); reading order can merge neighbouring columns (issue #5132) | Ollama and SGLang support were still pending official releases on the model card; no int4 VRAM figure published |
Sources
- Docling GitHub README
- Supported formats
- Issue #5254 OCR regression
- Issue #5132 reading order
- Hugging Face model card
- MiniCPM-V 4.5 paper
What each one is
MiniCPM-V 4.5
8B vision model from OpenBMB (August 2025) built to run on devices, with strong OCR and document parsing.
MiniCPM-V 4.5 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": ["openbmb/minicpm-v"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Docling alternativesMiniCPM-V 4.5 alternativesBest OCR models in 2026