MiniCPM-V 4.5 vs Unstructured
Pick MiniCPM-V 4.5 if you want to run OCR yourself on one GPU or on-device with quantized builds, and you only need text, Markdown or JSON back. Pick Unstructured if you ingest mixed files (PDF, Word, PowerPoint, email, HTML, images) for RAG and need typed elements with bounding boxes. Tables are the weak spot of the open-source Unstructured library, so use the hosted platform or another tool if table accuracy matters.
MiniCPM-V 4.5openbmb/minicpm-v | Unstructuredunstructured/unstructured | |
|---|---|---|
| Type | Vision model | OCR API |
| Price | 2 credits / page | 20 credits / page |
| Per 1,000 pages | $2.00 | $20.00 |
| Typical speed | 2.5 s / page | 3.0 s / page |
| Best for | OCR on edge devices | Mixed file types into RAG |
| Inputs | images | PDF, images, DOCX |
| Languages | 30+ languages | Multilingual |
| Max pages per request | 500 | 1,000 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | No | No |
| Bounding boxes | No | ✓ Yes |
| 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 MiniCPM-V 4.5 if
- It runs locally with llama.cpp or vLLM, with int4, GGUF and AWQ builds, so documents stay on your own hardware.
- Weights and code are Apache-2.0 and there is no per-page fee, only your own compute cost.
- OpenBMB claims leading OCRBench and OmniDocBench PDF parsing results among general multimodal models, though the model card shows scores only as images.
Choose Unstructured if
- It reads PDFs, HTML, Word, PowerPoint, emails and images in one library, so you do not need a separate parser per file type.
- It returns bounding boxes and typed elements as JSON or HTML, which MiniCPM-V 4.5 does not give you.
- The hosted API gives you 10,000 free pages, then $0.015 per page, with no GPU to run.
Key differences
| MiniCPM-V 4.5 | Unstructured | |
|---|---|---|
| How you run it | Self-host only: llama.cpp, vLLM, Ollama or Transformers. Ollama and SGLang support were still pending on the model card. | Self-host the Python library or call the hosted API. |
| Licence | Apache-2.0 weights and code | Apache-2.0 for the open-source library, proprietary for the hosted platform |
| Price | Free, you pay for compute | Hosted: first 10,000 pages free, then $0.015 per page. Business is custom. Library is free. |
| Output formats | Plain text, Markdown, JSON | JSON, HTML, plain text, bounding boxes |
| Layout and coordinates | General VLM with no bounding box or layout output | Returns bounding boxes and typed elements |
| Input file types | Not published | PDFs, HTML, Word, PowerPoint, emails and images |
| Known failure | Benchmark numbers are hard to verify because the model card gives them as images. No int4 VRAM figure is published. | Open-source table cell content scores 0.426 versus 0.866 hosted. hi_res tables cut off at the edges and drop first letters. Table OCR ignores language settings. |
Sources
- Hugging Face model card
- MiniCPM-V GitHub repo
- Unstructured pricing
- Unstructured GitHub README
- Unstructured issue #3262
- Unstructured issue #1694
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 detailsUnstructured
Document ETL for RAG. Partitions many file types into typed elements; open-source library too.
Unstructured 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": "openbmb/minicpm-v","fallbacks": ["unstructured/unstructured"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
MiniCPM-V 4.5 alternativesUnstructured alternativesBest OCR models in 2026