MinerU vs MiniCPM-V 4.5
Pick MinerU when you need to parse PDFs and Office files into structured Markdown, HTML or LaTeX, including on a CPU-only box. Pick MiniCPM-V 4.5 when you want a general vision model under a plain Apache-2.0 licence that you can quantize and run on one GPU or an edge device. The biggest gap is that MinerU is a document parsing toolkit, while MiniCPM-V 4.5 is an 8.7B general model with no layout output.
MinerUopendatalab/mineru | 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 | 2.0 s / page | 2.5 s / page |
| Best for | Academic PDFs | OCR on edge devices |
| Inputs | PDF, images | images |
| Languages | 80+ languages | 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 | ✓ Yes | ✓ 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 MinerU if
- You need one tool that accepts PDF, images, DOCX, PPTX, XLSX, EPUB, OFD and HTML and writes 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 documents are Chinese and English academic papers or mixed Office and PDF corpora for RAG.
Choose MiniCPM-V 4.5 if
- You need a plain permissive licence: both the weights and the code are Apache-2.0.
- You want to run quantized models on device or at the edge, using int4, GGUF or AWQ builds on llama.cpp or vLLM.
- You want local document Q&A on a single GPU, not only text extraction.
Key differences
| MinerU | MiniCPM-V 4.5 | |
|---|---|---|
| What it is | Document parsing toolkit with CLI, API and WebUI; MinerU2.5 VLM is 1.2B parameters | General 8.7B vision language model built on Qwen3-8B and SigLIP2-400M |
| Licence | MinerU Open Source License (Apache-2.0 based with extra conditions); MinerU2.5 VLM weights tagged AGPL-3.0 | Apache-2.0 |
| Input and output | Inputs: PDF, images, DOCX, PPTX, XLSX, EPUB, OFD, HTML. Outputs: Markdown, HTML, LaTeX, DOCX, EPUB, JSON | Outputs: plain text, Markdown, JSON. No bounding box or layout output |
| Hardware | Basic tier: CPU with 2 GB RAM. Standard and advanced VLM tiers: 16 GB RAM and a GPU with 8 GB+ VRAM | No int4 VRAM figure published |
| How you run it | Local CLI, API or WebUI. Version 4.1 changed HTTP APIs, so servers and clients must upgrade together | llama.cpp, vLLM, Ollama or Transformers. Ollama and SGLang support were still pending official releases on the model card |
| Benchmarks | olmOCR-Bench overall 75.2 (MinerU 2.5.4); ParseBench mean 45.9 (MinerU2.5) | Not published as text. OpenBMB claims leading OCRBench results and state of the art PDF parsing among general MLLMs on OmniDocBench, but the scores are only in images |
| Known failure | Repeated content in the VLM pipeline (300 KB figure blocks), dropped merged header rows in tables such as bank statements, and full-width Chinese punctuation converted to half-width | Benchmark numbers are hard to verify; no coordinates for downstream use |
Sources
- MinerU GitHub README
- MinerU2.5 model card
- olmOCR-Bench table
- Issue #5554 table header rows
- Hugging Face model card
- MiniCPM-V 4.5 paper
What each one is
MinerU
Open-source PDF to Markdown pipeline popular for academic and Chinese documents.
MinerU detailsMiniCPM-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": "opendatalab/mineru","fallbacks": ["openbmb/minicpm-v"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
MinerU alternativesMiniCPM-V 4.5 alternativesBest OCR models in 2026