Marker vs MiniCPM-V 4.5
Pick Marker to turn PDFs, books and papers with equations into Markdown, HTML or JSON, especially if you want RAG chunks or a CPU-only mode. Pick MiniCPM-V 4.5 for Apache-2.0 on-device OCR or document Q&A on a single GPU, where the licence has no revenue limit. The main split is licence and tooling: Marker's weights restrict commercial use above $5M, while MiniCPM-V is a general vision model with no layout or coordinate output.
Markerdatalab/marker | MiniCPM-V 4.5openbmb/minicpm-v | |
|---|---|---|
| Type | OCR API | Vision model |
| Price | 14 credits / page | 2 credits / page |
| Per 1,000 pages | $14.00 | $2.00 |
| Typical speed | 2.5 s / page | 2.5 s / page |
| Best for | Books and research papers | OCR on edge devices |
| Inputs | PDF, images, DOCX | images |
| Languages | 90+ languages | 30+ languages |
| Max pages per request | 1,000 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | ✓ Yes |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | No |
| 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 Marker if
- Your company is under $5M or you only need research or personal use, and you want a PDF to Markdown pipeline with no per-page cost.
- You need structured output: Markdown, HTML, JSON block trees, LaTeX or RAG chunks.
- You have no GPU: fast mode runs on CPU at 23.6 pages per second with about 18 MB of ONNX weights, though it scores 66.6 on olmOCR-bench.
Choose MiniCPM-V 4.5 if
- You need Apache-2.0 weights and code with no revenue cap on commercial use.
- You want quantized on-device or edge OCR, with int4, GGUF and AWQ builds that run on llama.cpp and vLLM.
- You want to ask questions about a document locally on one GPU, not only convert it.
Key differences
| Marker | MiniCPM-V 4.5 | |
|---|---|---|
| Licence | Code Apache-2.0; weights under a modified AI Pubs Open Rail-M license, free for research, personal use and startups under $5M | Apache-2.0 for weights and code |
| What it is | PDF to Markdown converter built on the Surya OCR models | 8.7B general vision language model built on Qwen3-8B and SigLIP2-400M |
| How you run it | Locally on GPU, CPU or Apple Silicon; Datalab also sells a hosted API | Self-host only, with llama.cpp, vLLM or Transformers; Ollama and SGLang support were still pending on the model card |
| Price | Free self-hosted within the weight license; hosted API includes about 5,000 free pages per month | Free self-hosted; you pay for compute |
| Output formats | Markdown, HTML, JSON, chunks, LaTeX, images | Plain text, Markdown, JSON; no bounding box or layout output |
| Benchmarks | 83.7 on olmOCR-bench and 85.9 on OmniParseBench in vllm mode, both reported by Datalab | Claims leading OCRBench results and state of the art PDF parsing on OmniDocBench, but scores are only published as images |
| Hosted option | Datalab hosted API with paid plans | Not published |
Sources
What each one is
Marker
Hosted version of the open-source Marker and Surya models. Strong on books and papers.
Marker detailsMiniCPM-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 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": "datalab/marker","fallbacks": ["openbmb/minicpm-v"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Marker alternativesMiniCPM-V 4.5 alternativesBest OCR models in 2026