OCR models / OpenDataLab (Shanghai AI Lab)
MinerU2.5 (2509)
OpenDataLab (Shanghai AI Lab) · released 2025-09-26 · data checked 2026-11-06
MinerU2.5 (2509) is a 1.2B-parameter OCR and document-parsing model from OpenDataLab (Shanghai AI Lab) (CN), released 2025-09-26 under the AGPL-3.0 license. It scores 93.04 on OmniDocBench v1.6 (#21 of 33) and 75.2 on olmOCR-Bench (#20 of 24). It needs about 3.3 GB of memory at 16-bit.
The original MinerU2.5, succeeded by MinerU2.5 Pro. PDF to Markdown with the MinerU toolkit.
OmniDocBench v1.6
93.04
#21 of 33 models
olmOCR-Bench
75.2
#20 of 24 models
Parameters
1.2B
OCR specialist
Memory to run
~3.3 GB
16-bit weights plus about 1 GB
Specifications
- Lab
- OpenDataLab (Shanghai AI Lab) (CN)
- Type
- OCR / document parsing specialist
- Total parameters
- 1.2B
- License
- AGPL-3.0AGPL-3.0: if you offer it as a network service, you must publish your changes under the same license.
- Memory to run
- ~3.3 GB at 16-bit
- Released
- 2025-09-26
Closest alternatives
- FireRed-OCR (2.1B, 93.26 on OmniDocBench, Apache-2.0)
- ABot-OCR (2.1B, 93.30 on OmniDocBench, Not stated)
- Logics-Parsing v2 (4.4B, 93.33 on OmniDocBench, Not stated)
- Ovis2.6 30B-A3B (31.4B, 93.70 on OmniDocBench, Apache-2.0)
Sources
- https://huggingface.co/opendatalab/MinerU2.5-2509-1.2B
- https://github.com/opendatalab/OmniDocBench
- https://arxiv.org/abs/2509.22186
Something wrong here? Report it.