OCRRouter

Qianfan OCR vs MiniCPM-V 4.5

Pick Qianfan OCR for self-hosted document parsing where you need layout output, 192 languages, handwriting, tables and formulas, plus a published 93.12 on OmniDocBench v1.5. Pick MiniCPM-V 4.5 for on-device or single-GPU OCR and document Q&A with quantized builds, as long as you do not need coordinates.

Qianfan OCRbaidu/qianfan-ocrMiniCPM-V 4.5openbmb/minicpm-v
TypeOCR APIVision model
Price2 credits / page2 credits / page
Per 1,000 pages$2.00$2.00
Typical speed2.0 s / page2.5 s / page
Best forChinese documentsOCR on edge devices
InputsPDF, imagesimages
LanguagesChinese, English and more30+ languages
Max pages per request200500
Tables✓ Yes✓ Yes
Handwriting✓ Yes✓ Yes
Math✓ YesNo
Bounding boxesNoNo
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 Qianfan OCR if

  • It returns Markdown, JSON, HTML, LaTeX and bounding boxes, so you get layout and coordinates.
  • It scores 93.12 on OmniDocBench v1.5, 79.8 on olmOCR-Bench and 880 on OCRBench, with numbers published on the model card.
  • It covers 192 languages, handwriting and scene text, and suits key information extraction, tables and formulas.

Choose MiniCPM-V 4.5 if

  • It runs on llama.cpp and vLLM with int4, GGUF and AWQ builds, which fits edge and on-device OCR.
  • It handles local document Q&A on a single GPU, because it is a general vision language model and not only a parser.
  • It comes with Apache-2.0 weights and code, and OpenBMB claims leading OCRBench results above GPT-4o-latest and Gemini 2.5.

Key differences

Qianfan OCRMiniCPM-V 4.5
Model size4B, built on Qwen3-4B8.7B, built on Qwen3-8B and SigLIP2-400M
LicenceApache-2.0Apache-2.0
How you run itSelf-host the open weights, or use the Baidu Qianfan platformSelf-host only, with llama.cpp, vLLM or Transformers; Ollama and SGLang support were still pending
Output formatsMarkdown, JSON, HTML, LaTeX, bounding boxesPlain text, Markdown, JSON; no bounding box or layout output
BenchmarksOmniDocBench v1.5 93.12, olmOCR-Bench 79.8, OCRBench 880Not published as text; the model card shows scores only as images
HardwareBenchmarked on one NVIDIA A100 with vLLM; no minimum VRAM stated; not for CPU-only useInt4, GGUF and AWQ builds exist; no int4 VRAM figure published
Known weaknessesThinking mode can emit up to 16,384 tokens, adding latency; 1.024 pages per second on an A100 even with W8A8Benchmark numbers are hard to verify; not suited to jobs needing coordinates or a hosted SLA

Sources

What each one is

Qianfan OCR

Baidu's OCR-tuned vision model. Strong on Chinese documents and tables.

Qianfan 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 10 files at onceNo file?

Use one, fall back to the other

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

Questions

More comparisons

Qianfan OCR alternativesMiniCPM-V 4.5 alternativesBest OCR models in 2026