OCRRouter

DeepSeek-OCR vs MiniCPM-V 4.5

Pick DeepSeek-OCR for cheap, self-hosted bulk OCR where you want Markdown, HTML and bounding boxes under an MIT licence. Pick MiniCPM-V 4.5 for on-device or single-GPU document Q&A with quantized builds and JSON output. The main split is that DeepSeek-OCR returns coordinates, while MiniCPM-V 4.5 has no bounding box or layout output.

DeepSeek-OCRdeepseek/deepseek-ocrMiniCPM-V 4.5openbmb/minicpm-v
TypeOpen sourceVision model
Price1 credit / page2 credits / page
Per 1,000 pages$1.00$2.00
Typical speed1.1 s / page2.5 s / page
Best forCheap bulk conversionOCR on edge devices
InputsPDF, imagesimages
Languages100+ languages30+ languages
Max pages per request200500
Tables✓ Yes✓ Yes
HandwritingNo✓ Yes
Math✓ YesNo
Bounding boxes✓ YesNo
ConfidenceNoNo
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 DeepSeek-OCR if

  • You need bounding boxes, since it outputs them alongside Markdown, plain text and HTML.
  • You run bulk jobs on a CUDA GPU: the authors quote about 2,500 tokens per second PDF throughput on one A100-40G.
  • You want to trade accuracy for cost per page, because resolution modes run from 64 to 400 vision tokens.

Choose MiniCPM-V 4.5 if

  • You want to run on small hardware, as llama.cpp and vLLM builds exist in int4, GGUF and AWQ.
  • You want JSON output or to ask questions about a document locally on a single GPU.
  • You prefer Apache-2.0 weights and code, and a general 8.7B vision language model rather than an OCR-only one.

Key differences

DeepSeek-OCRMiniCPM-V 4.5
LicenceMITApache-2.0
How you run itSelf-host with Transformers or vLLM; reference setup is CUDA 11.8 and PyTorch 2.6 in bfloat16Self-host with llama.cpp, vLLM, Ollama or Transformers; Ollama and SGLang support were still pending official releases on the model card
PriceFree, you pay for computeFree, you pay for compute
Output formatsMarkdown, plain text, HTML, bounding boxesPlain text, Markdown, JSON; no bounding box or layout output
Model typeOCR model that compresses pages into few vision tokensGeneral 8.7B vision language model built on Qwen3-8B and SigLIP2-400M
BenchmarksNot publishedOpenBMB claims leading OCRBench results and state of the art PDF parsing among general MLLMs on OmniDocBench; scores are given only as images, so they are hard to verify
Known failureRepeated tokens and looping over table headers (GitHub issues #12 and #16); corrects nonsense words instead of copying themNo int4 VRAM figure published

Sources

What each one is

DeepSeek-OCR

Open-source vision model that compresses pages into few tokens.

DeepSeek-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": "deepseek/deepseek-ocr","fallbacks": ["openbmb/minicpm-v"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

DeepSeek-OCR alternativesMiniCPM-V 4.5 alternativesBest OCR models in 2026