🚀 Introducing MinerU Document Explorer!
Inspired by Karpathy’s LLM Wiki vision — build your dynamic, self-maintaining AI knowledge base beyond RAG.
Parse PDFs/Word/PPT, auto-generate linked Wiki, precise extraction & full traceability.
Lightweight, local, agent-native.
🚀MinerU2.5-Pro!
Same 1.2B architecture, NO structural changes.
65.5M pages, diversity-aware sampling, cross-model verification, render-then-verify correction.
Scores 95.69 on OmniDocBench v1.6, outperforming larger models.
Built for real-world use
Now open-sourced!
#MinerU
Struggling with messy receipts & invoices? 🧾
#MinerU Skills delivers zero-code receipt parsing.
Automatically locate, split & extract key data—amounts, dates, items—with high accuracy.
Structured output ready for storage & reconciliation.
Full workflow tutorial video now live!
Tired of messy PDF outputs? 📄
#MinerU Skills let you process papers with zero code.
Parse layouts, formulas, tables & OCR in one click.
Batch 50+ papers to clean Markdown, keep LaTeX & tables perfectly.
Connect to RAG/knowledge base effortlessly.
Full tutorial video out now!
🚀#OpenDataLab’s AI-ready database #Sciverse Launched!
Powering #AGI4S with a 3-layer system (Sci-Base/Sci-Align/Sci-Evo).
✅25M+ parsed literatures, 600B+ high-quality tokens via #MinerU
✅18M+ protein sequences, 6M+ chemical reactions
👉Explore: https://t.co/Jrqa8mwXo0
🚀New from #OpenDataLab: MinerU-Diffusion!
We redefine document OCR as inverse rendering via diffusion decoding, replacing slow autoregressive generation.
✅ Up to 5.1× faster inference
✅ Stronger visual structure modeling
✅ Stable in challenging scenarios
Try it & star us!
🚀Big Update! MinerU has adapted to 10+ computing power platforms
💯99% accuracy in capturing PDF/web elements
💪OmniDocBench adopted by Gemini3/DeepSeek as authoritativee benchmark
👉Explore: https://t.co/kQPNkWoiPr;
https://t.co/7BCa9fJlq6
🏆MDIC: https://t.co/wbqIBVESe0
MinerU2.5 is a compact 1.2B VLM with a smart two-stage, coarse-to-fine pipeline (global layout → native-res crops) that delivers state-of-the-art doc parsing with low compute.