MinerU2.5-Pro

by OpenDataLab (Shanghai AI Laboratory)

MinerU2.5-Pro is a 1.2-billion-parameter vision-language model specialized for high-performance document parsing and understanding. It achieves state-of-the-art results on the OmniDocBench v1.6 benchmark (95.69 overall score) through data-centric engineering rather than architectural changes, processing complex layouts with text, tables, formulas, and images.

Vision Models
Proprietary

Primary Use Cases

  • Enterprise document intelligence and parsing
  • Academic paper and research document extraction
  • Complex table and formula recognition
  • RAG pipeline document preprocessing
  • Legal and financial document processing

✅ Strengths

  • State-of-the-art on OmniDocBench v1.6 (95.69 score)
  • Handles complex nested tables, dense formulas, and multi-column layouts
  • Data-centric approach: improvements without architectural changes
  • Efficient 1.2B parameters — deployable on modest hardware
  • Open-source with community quantizations (GGUF, MLX)

⚠️ Limitations

  • Specialized for document parsing only — not a general-purpose model
  • 1.2B parameters limits handling of extremely complex reasoning tasks
  • Requires vllm-async-engine for high-concurrency production inference
  • Training data focused on academic and structured documents

🏆 Performance Benchmarks

Standardized benchmark scores for objective comparison

OmniDocBench v1.6(Document Understanding)
95.69overall score

State-of-the-art on OmniDocBench v1.6 protocol. CDM (formula recognition) score: 97.29.

Technical Specifications

Developer
OpenDataLab (Shanghai AI Laboratory)
Category
Vision Models
Type
🧠 Foundation Model
License
UNKNOWN

Performance Benchmarks

OmniDocBench v1.695.69overall score

Source: View benchmark details

Trust & Privacy

Health Status
🟢 Active
Privacy Grade
🛡️ Grade A
Trains on User Data
✅ Does not train
Certifications
None listed

Ecosystem & Integrations

API Access
❌ No
Chrome Extension
❌ No
Mobile App
❌ No
Pricing Model
🔓 Open Source
Free Tier
✅ Available

Pricing Breakdown

Open Source

Free to download and self-host from Hugging Face

0