MinerU 2.5
MinerU 2.5 is OpenDataLab's 1.2B document parsing model. It first finds the layout on a small copy of the page, then reads each region at full resolution.
A release of MinerU.
- Released
- September 2025
- Model ID
- opendatalab/MinerU2.5-2509-1.2B
- Size
- 1.2B vision-language model
- License
- AGPL-3.0
- Also searched as
- mineru 2.5, mineru2.5, MinerU2.5-2509-1.2B
What is new in MinerU 2.5
- Two-stage parsing: layout analysis on a downsampled page, then recognition on full-resolution crops.
- Finer layout analysis that keeps headers, footers and page numbers.
- Better formulas, including long ones and equations that mix Chinese and English.
- More robust on rotated, borderless and partly bordered tables.
MinerU 2.5 benchmarks
| Benchmark | Score | Note |
|---|---|---|
| OmniDocBench, overall | 90.67 | 1.82 above MonkeyOCR-pro-3B, 2.26 above dots.ocr |
| OmniDocBench, text edit distance | 0.047 | dots.ocr: 0.048. Lower is better |
| OmniDocBench, formula CDM | 88.46 | |
| Throughput | 2.12 pages/s | vLLM async engine, one A100 |
Scores as published by the authors. We have not rerun them.
Use MinerU 2.5 through an API
OCRRouter runs a different release of MinerU as opendatalab/mineru.
1DeepSeek-OCROpen-source vision model that compresses pages into few tokens.1 credit / page
deepseek/deepseek-ocr1 credit / page2PaddleOCR-VLOpen-source document model from Baidu. Excellent on Chinese, Japanese and Korean.6 credits / pagepaddle/paddleocr-vl6 credits / page3GLM-OCRSmall open-source OCR model from Z.ai with layout regions. MIT licensed.6 credits / pagezai/glm-ocr6 credits / pageMinerU versions
- MinerU: all versions, price, speed and features on OCRRouter
- MinerU 2.5 (September 2025), this page
Questions
Sources
Last updated .