GOT-OCR 2.0 vs MiniCPM-V 4.5
Pick GOT-OCR 2.0 for formulas, sheet music, chemical structures and region-specific OCR on English and Chinese documents. Pick MiniCPM-V 4.5 when you want a general vision language model with an Apache-2.0 licence on the weights and code, quantized builds for a single GPU or edge device, and document Q&A. The main split is scope: GOT-OCR 2.0 is a 580M specialist with special output formats, MiniCPM-V 4.5 is an 8.7B general model.
GOT-OCR 2.0stepfun/got-ocr-2 | 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 | 1.5 s / page | 2.5 s / page |
| Best for | Formatted text, formulas and charts | OCR on edge devices |
| Inputs | images | images |
| Languages | Chinese, English | 30+ languages |
| Max pages per request | 100 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | ✓ Yes |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | No |
| Confidence | No | No |
| Forms | No | 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 GOT-OCR 2.0 if
- You need LaTeX, TikZ, SMILES or kern output for formulas, chemical structures or sheet music.
- You want to OCR one region of a page, guided by box coordinates or color.
- You want a small 580M model with multi-page OCR and batched inference in Hugging Face Transformers.
Choose MiniCPM-V 4.5 if
- You want Apache-2.0 on both weights and code, with no non-commercial training data licence to check.
- You need quantized builds (int4, GGUF, AWQ) that run on llama.cpp or vLLM for on-device or single-GPU use.
- You want to ask questions about documents locally and get Markdown or JSON back, not only transcribe them.
Key differences
| GOT-OCR 2.0 | MiniCPM-V 4.5 | |
|---|---|---|
| Licence | Code Apache-2.0; training data CC BY-NC 4.0 | Apache-2.0 |
| Model size | 580M parameters | 8.7B parameters, built on Qwen3-8B and SigLIP2-400M |
| How you run it | Self-hosted with Hugging Face Transformers. README lists CUDA 11.8 and PyTorch 2.0.1. CPU users are pointed to a llama.cpp port. | Self-hosted with llama.cpp, vLLM or Transformers. Int4, GGUF and AWQ builds exist. Ollama and SGLang support were still pending on the model card. |
| Output formats | Plain text, Markdown, LaTeX, TikZ, SMILES, kern | Plain text, Markdown, JSON |
| Region and coordinates | Region OCR guided by box coordinates or color | No bounding box or layout output |
| Benchmarks | Not published in the README | OpenBMB claims leading OCRBench and OmniDocBench PDF parsing results, but scores are given only as images |
| Known failure modes | Loops characters midway through recognition (issue #226). Returns wrong text on images with no text (issue #235). Format mode is much worse than plain OCR mode (issue #317). | No int4 VRAM figure published. No hosted SLA. |
Sources
What each one is
GOT-OCR 2.0
Unified OCR model that reads plain text, tables, formulas, sheet music and charts.
GOT-OCR 2.0 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": "stepfun/got-ocr-2","fallbacks": ["openbmb/minicpm-v"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
GOT-OCR 2.0 alternativesMiniCPM-V 4.5 alternativesBest OCR models in 2026