OCRRouter

Pixtral vs MiniCPM-V 4.5

Pick MiniCPM-V 4.5 for new self-hosted OCR work, such as on-device or single-GPU document Q&A. Pixtral 12B is deprecated and no longer maintained by Mistral, so keep it only for existing Mistral API integrations. Neither model returns coordinates or layout output.

Pixtralmistral/pixtralMiniCPM-V 4.5openbmb/minicpm-v
TypeVision modelVision model
Price2 credits / page2 credits / page
Per 1,000 pages$2.00$2.00
Typical speed3.0 s / page2.5 s / page
Best forQuestions about document imagesOCR on edge devices
Inputsimagesimages
LanguagesMultilingual30+ languages
Max pages per request500500
TablesNo✓ Yes
HandwritingNo✓ Yes
MathNoNo
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 Pixtral if

  • You already run on the Mistral API and want to keep the pixtral-12b-2409 integration, which launched at $0.15 per million input and output tokens.
  • You need question answering over charts and documents, where Mistral reports DocVQA 90.7 ANLS and ChartQA 81.8 (CoT).
  • You send many images in one request, since the 128K token context allows it.

Choose MiniCPM-V 4.5 if

  • You are starting a new project and want a model that is still being developed, rather than one Mistral marks as deprecated.
  • You need to run on-device or at the edge, with int4, GGUF and AWQ builds on llama.cpp (b6282) or vLLM (v0.10.2).
  • You want an 8.7B model on a single GPU for local document Q&A, with OpenBMB claiming leading OCRBench results and state of the art PDF parsing on OmniDocBench.

Key differences

PixtralMiniCPM-V 4.5
StatusDeprecated by Mistral, no longer maintainedCurrent OpenBMB release
How you run itMistral API (pixtral-12b-2409) or open weightsSelf-host only, with llama.cpp, vLLM, Ollama or Transformers
LicenceApache-2.0Apache-2.0 weights and code
Price$0.15 per million input and output tokens at API launch; free to self-host plus computeFree, self-hosted, plus compute
Size12.4B (12B decoder + 400M vision encoder)8.7B (Qwen3-8B + SigLIP2-400M)
Published benchmarksDocVQA 90.7 ANLS, ChartQA 81.8, MMMU 52.5 (reported by Mistral)Not published as text. The model card shows scores only as images, so they are hard to verify
Known OCR failuresReported drops on handwriting, small fonts and non-Latin scripts. Q4 quantization degrades small text OCRNo bounding box or layout output. Ollama and SGLang support were still pending on the model card. No int4 VRAM figure published

Sources

What each one is

Pixtral

Mistral's open-weight vision models, Pixtral 12B and Pixtral Large. Mistral's API no longer serves them, so they run self-hosted.

Pixtral 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": "mistral/pixtral","fallbacks": ["openbmb/minicpm-v"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Pixtral alternativesMiniCPM-V 4.5 alternativesBest OCR models in 2026