markerDatalab's 39k-star PDF-to-Markdown/JSON converter: a layout pipeline plus an optional LLM pass for tables, forms and equations, with chunk output and form-value extraction built in.
Why switcholmOCR runs one vision-language model over every page; marker runs layout analysis first and calls a model only where needed — cheaper at corpus scale, weaker on truly messy scans.
Full comparison → MinerUHeavyweight document-to-markdown/JSON parser — PDFs plus Office (docx/pptx/xlsx) through layout analysis and OCR into LLM-ready output for RAG and agentic pipelines. 73k stars, self-hostable.
Why switchSame job — self-hosted document→markdown for LLM ingestion: olmocr is a focused VLM PDF-linearizer; MinerU is the broader pipeline (PDF + Office, layout analysis) with far more traction.
Full comparison → Unlimited-OCRBaidu's open OCR VLM that parses entire multi-page documents in one shot — 'unlimited' long-horizon parsing pushing DeepSeek-OCR further. MIT weights on HF; serve via transformers, vLLM or SGLang.
Why switchSame job — self-hosted VLM that linearizes messy PDFs into LLM-ready text: olmocr is AllenAI's page-pipeline toolkit built for training-data ingestion; Unlimited-OCR bets on one-shot long-horizon parsing that preserves cross-page structure.
Full comparison → chandraDatalab's SOTA open OCR model: images/PDFs to structured HTML/Markdown/JSON with layout, tables, forms, checkboxes, handwriting and math, in 90+ languages. Local HF or vLLM inference.
Why switchHead-to-head open OCR-VLM rivals on the same benchmark (Chandra 2 scores 85.8 vs olmOCR 2's 82.4). olmOCR is fully permissive and tuned for LLM-training-data linearization; Chandra leads on handwriting/forms/multilingual but carries an OpenRAIL-M weights license.
Full comparison → xbergRust-core document-intelligence engine with 15 language bindings: turns 96 formats — PDF, Office, images, audio, code — into clean text, tables and RAG-ready chunks. Library, CLI, REST or MCP.
Why switchSame end goal — clean, ordered Markdown from PDFs for RAG/training ingestion. olmocr is a dedicated self-hosted vision-language model that excels on scans, tables and handwriting; Xberg is a general extraction framework where OCR is one pluggable backend. Use olmocr when OCR quality is the bottleneck, Xberg when you need many formats and language bindings.
Full comparison → TurboOCRGPU document parser in C++/TensorRT: PP-OCRv6 OCR, layout, tables and formulas to Markdown at 200-559 img/s on one GPU — no VLM, HTTP + gRPC, one-line Docker deploy.
Why switcholmOCR linearizes messy PDFs with a self-hosted VLM; TurboOCR refuses the VLM entirely and wins ~20x on throughput. Pick by corpus: pristine forms → TurboOCR, chaotic scans → VLM.
Full comparison → chunkrDocument intelligence API in Rust: layout analysis, OCR with bounding boxes, and semantic chunking that turn PDFs, PPTs and Word docs into RAG-ready structured chunks.
Why switchBoth parse hard documents for AI consumption; olmocr bets on a VLM end-to-end, Chunkr on a layout-analysis pipeline with structured outputs and citations.
Full comparison → PixelRAGBerkeley's visual RAG: render pages and PDFs to screenshot tiles and retrieve with a VLM embedder — tables, charts and layout survive. pixelshot CLI plus a hosted 8.28M-page Wikipedia index.
Why switcholmocr linearizes messy PDFs into clean ordered Markdown so a text pipeline can index them; PixelRAG argues the linearization step is the loss — embed the page image and let a VLM read the tile directly.
Full comparison → doctrdocTR: two-stage OCR in PyTorch — detect words, then recognize them — with pretrained detection and recognition architectures you can mix, plus layout detection and rotated-page handling.
Why switchBoth turn messy pages into text for downstream AI. olmOCR linearizes whole documents into ordered Markdown with a self-hosted VLM, aimed at training data and RAG ingestion; docTR stops at accurate word-level recognition you can build any layout logic on.
Full comparison →