MinerU

Based on the available evidence, MinerU looks like a broadly adopted and seriously evaluated document parsing project, and it is worth shortlisting first if your workflow depends on complex documents. Still, it is better understood as structured extraction infrastructure than as a plug-and-play OCR app. GitHub stars, Trending visibility, and repost-heavy X posts are strong proof of attention, not proof of quality by themselves. The more useful signals for capability come from the official repository, a Zhihu hands-on usage answer, and tutorial-style content showing export and translation workflows. My adoption take: promising and practical for LLM knowledge-base and RAG preprocessing, but non-technical teams should validate with their own sample set.

What MinerU actually does is more than basic text recognition. It tries to preserve layout, formulas, chart positions, and document structure from PDFs and Office files, then converts them into Markdown or JSON for downstream use. A more accurate analogy is an LLM-oriented document parsing pipeline, not a generic scanner OCR product.