Donut vs MiniCPM-V 4.5
Pick MiniCPM-V 4.5 if you want OCR that works without training, including Markdown output and document Q&A on a single GPU or edge device. Pick Donut if you have a narrow, fixed-layout document type such as receipts or forms and are willing to fine-tune it for structured JSON. The main split is that Donut is a pretrain-only base model that needs training, while MiniCPM-V 4.5 is a general vision language model you can run as is.
Donutnaver/donut | MiniCPM-V 4.5openbmb/minicpm-v | |
|---|---|---|
| Type | Open source | Vision model |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 1.0 s / page | 2.5 s / page |
| Best for | Receipt and form fields after fine-tuning | OCR on edge devices |
| Inputs | images | images |
| Languages | English, Chinese, Japanese, Korean | 30+ languages |
| Max pages per request | 100 | 500 |
| Tables | No | ✓ Yes |
| Handwriting | No | ✓ Yes |
| Math | No | No |
| Bounding boxes | No | No |
| Confidence | No | No |
| Forms | ✓ Yes | No |
| Multilingual | No | ✓ 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 Donut if
- You process one fixed-layout document type, such as receipts or forms, and can fine-tune on it.
- You want structured JSON straight from the image with no separate OCR step.
- You need document classification, which is a listed use for Donut.
Choose MiniCPM-V 4.5 if
- You need OCR or document Q&A without any training run first.
- You want Markdown or JSON output from full pages, and OpenBMB claims state of the art PDF parsing among general MLLMs on OmniDocBench.
- You want quantized on-device or edge deployment, with int4, GGUF and AWQ builds on llama.cpp and vLLM.
Key differences
| Donut | MiniCPM-V 4.5 | |
|---|---|---|
| Ready to use | No. The base checkpoint is pretrain-only and must be fine-tuned first. | Yes. It is a general vision language model you can run directly. |
| Licence | MIT | Apache-2.0 for weights and code |
| Output formats | JSON, plain text | Plain text, Markdown, JSON |
| How you run it | Self-host. Available in Hugging Face Transformers, but Transformers v5 dropped image-to-text pipeline support, so load it directly or use v4. | Self-host with llama.cpp, vLLM, Ollama or Transformers. Ollama and SGLang support were still pending official releases on the model card. |
| Price | Free, self-hosted (you pay for compute) | Free, self-hosted (you pay for compute) |
| Small text on large pages | Misses tiny text in large images because of input resolution limits. | Not published |
| Benchmarks | Not published. A Hacker News user found results on personal documents did not match the promised scores. | OpenBMB claims leading OCRBench results above GPT-4o-latest and Gemini 2.5, but the model card shows scores only as images, so they are hard to verify. |
Sources
What each one is
Donut
OCR-free document understanding model. Fine-tuned for receipts, forms and document Q&A.
Donut 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": "naver/donut","fallbacks": ["openbmb/minicpm-v"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Donut alternativesMiniCPM-V 4.5 alternativesBest OCR models in 2026