OCRRouter

PaddleOCR (PP-OCR) vs MiniCPM-V 4.5

Pick PaddleOCR (PP-OCR) for fast text lines and boxes on printed Chinese and English, receipts, signs and screenshots, especially on CPU, edge devices or in the browser. Pick MiniCPM-V 4.5 when you want Markdown or JSON from a document and can run an 8.7B model on a GPU. The main split is a 34.5M-parameter detector and recognizer that returns coordinates, against a general vision language model that returns no coordinates.

PaddleOCR (PP-OCR)paddle/paddleocrMiniCPM-V 4.5openbmb/minicpm-v
TypeOpen sourceVision model
Price1 credit / page2 credits / page
Per 1,000 pages$1.00$2.00
Typical speed0.7 s / page2.5 s / page
Best forFast printed text, Chinese and EnglishOCR on edge devices
InputsPDF, imagesimages
Languages80+ languages30+ languages
Max pages per request500500
TablesNo✓ Yes
HandwritingNo✓ Yes
MathNoNo
Bounding boxes✓ YesNo
Confidence✓ YesNo
FormsNoNo
Multilingual✓ Yes✓ Yes
On OCRRouterAvailableAvailable

Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.

Which to choose

Choose PaddleOCR (PP-OCR) if

  • It returns plain text, bounding boxes and JSON, so you get coordinates for every text line.
  • The PP-OCRv6 medium model has 34.5M parameters and runs on CPU, NVIDIA GPU and Kunlunxin XPU, with 0.13s reported on an A100 and a 5.2x CPU speedup with OpenVINO.
  • One PP-OCRv6 model covers 50 languages, including Chinese, English, Japanese and 46 Latin-script languages, and the official PaddleOCR.js SDK runs it in the browser.

Choose MiniCPM-V 4.5 if

  • It outputs Markdown and JSON, which suits document to text jobs that PP-OCR alone does not cover.
  • OpenBMB claims leading OCRBench results above GPT-4o-latest and Gemini 2.5, and state of the art PDF parsing among general MLLMs on OmniDocBench.
  • It runs on llama.cpp and vLLM with int4, GGUF and AWQ builds, which fits local document Q&A on a single GPU or on-device OCR with quantized models.

Key differences

PaddleOCR (PP-OCR)MiniCPM-V 4.5
Model typeText detection and recognition pipeline, 34.5M parameters (PP-OCRv6 medium)Vision language model, 8.7B parameters, built on Qwen3-8B and SigLIP2-400M
Licence and priceApache-2.0, free, self-hosted (you pay for compute)Apache-2.0 weights and code, free, self-hosted (you pay for compute)
OutputPlain text, bounding boxes, JSONPlain text, Markdown, JSON; no bounding box or layout output
How you run itpaddleocr Python package, C++ or the PaddleOCR.js browser SDKllama.cpp, vLLM, Ollama or Transformers; Ollama and SGLang support were still pending official releases on the model card
HardwareCPU, NVIDIA GPU, Kunlunxin XPU and other acceleratorsSingle GPU for local use; no int4 VRAM figure published
Tables and formulasNot handled by the base model; needs the separate PP-Structure or VL pipelinesNot published
Known issuesNotImplementedError on PaddleOCR 3.7.0 with PaddlePaddle 3.3.1 (issue #18162); PP-OCRv6 weights contain denormals that make CPU inference 4 to 11x slower in RapidOCR (issue #743)Benchmark scores on the model card are images only, so the numbers are hard to verify

Sources

What each one is

PaddleOCR (PP-OCR)

Baidu's classic open-source text detection and recognition pipeline. Light enough to run on CPU.

PaddleOCR (PP-OCR) details

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 details

Try both on your own file

Run the file once with each model and compare the output.

PDF, PNG, JPG, WebP, HEIC. Up to 20 MB each, 10 files.No file?

Use one, fall back to the other

{"model": "paddle/paddleocr","fallbacks": ["openbmb/minicpm-v"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

PaddleOCR (PP-OCR) alternativesMiniCPM-V 4.5 alternativesBest OCR models in 2026