Marker is Datalab's open source AI data processing tool. It can convert PDF and various documents into Markdown and JSON with high precision. It is often used to prepare structured text for large models and RAG pipelines.

Marker

Core parameters and statistics

Marker is Datalab's open source document conversion tool. The GitHub repository describes it as "Convert PDF to markdown + JSON quickly with high accuracy". It converts unstructured PDF and multi-format documents into structured text and is a common pre-processing tool for preparing corpus for large models and RAG (Retrieval Enhanced Generation) pipelines.

Projects	Public Information
Official positioning	Convert PDF/documents to Markdown and JSON with high precision
Core capabilities	Layout restoration, table recognition, formula and code block processing, multi-format output
Output format	Markdown, JSON, HTML
Deployment form	Open source command line/Python library, and Datalab cloud API
Open source license	GitHub public repository, open source can be self-hosted
Community size	About 36k stars, 2k forks
Maintainer	Datalab (datalab.to)

Positioning Difference: Marker is not a reader or editor, but a "document to structured text" conversion engine. Its value lies in retaining the original layout semantics (title hierarchy, tables, formulas) as much as possible, so that the conversion results can be directly fed to the search and question and answer system.

Community verification: The scale of about 36k stars and 2k forks shows that it has been widely used in RAG and document digitization scenarios, and has strong external verification for problem feedback and boundary cases.

User and market recognition

Marker's recognition mainly comes from the open source community and AI engineering practices. The official number of commercial users and revenue are not disclosed. These dimensions are not disclosed.

Community Popularity: GitHub public data shows about 36k stars and 2k forks. It is at the top of the open source document parsing tools and is often compared with other PDF parsing solutions.

Typical adopters: Most of the people who use Marker are development teams building RAGs, knowledge bases or document automation pipelines. They need to convert a large number of PDFs (papers, contracts, manuals, reports) into searchable text.

Prerequisites for implementation: To achieve value, certain engineering capabilities are required - installing dependencies, configuring models, and adjusting parameters according to document types. For users who don’t have R&D resources and just want to click and upload, Datalab’s cloud API is more suitable.

Cost advantage

The cost value of Marker lies in "open source and free + self-hosting", which reduces the marginal cost of large-volume document conversion to the infrastructure level.

C-side/Individual: Open source and free, can be run locally; the main cost for individuals is local computing power (some modes rely on GPU speed-up).

Developer/API: It can be integrated into the data pipeline as a Python library or command line, with zero licensing fee; if you do not want to build it yourself, you can use the cloud API provided by Datalab instead, and it will be billed based on the number of calls. The specific price is subject to the official page.

Enterprise/Private: The open source feature makes it easy to process sensitive documents (contracts, financial reports) on the intranet and in an isolated environment, avoiding uploading confidential PDFs to third parties. The true cost should be superimposed on GPU resources, operation and maintenance, and parameter adjustment investment by document type.

Cost structure tip: Compared with commercial document parsing API, self-hosted Marker eliminates the need for per-page billing, but increases computing power and maintenance costs. The larger the batch size and the more sensitive the documents, the more obvious the relative advantages of self-hosting become.

Main functions

Marker's capabilities are built around "converting documents into structured text with high fidelity":

• PDF to Markdown/JSON: Preserve title hierarchy, paragraph and list structure.
• Table recognition: Try to restore the row and column structure of the table instead of compressing it into plain text.
• Formula and code processing: Special processing of formulas and code blocks in academic and technical documents.
• Multi-format input: In addition to PDF, it supports more document formats and uniformly outputs structured text.
• LLM enhanced mode: Large models can be used to assist in improving the restoration accuracy of complex layouts and tables.

Acceptance concerns: Actual performance depends on source PDF quality (scan vs. native text), table complexity, and multi-column layout. It is recommended to use real document samples to evaluate the table and formula restoration rates before deciding whether to enable LLM enhancement.

Model and version evolution

The current public version information has been covered in the previous article. If the official does not fully disclose the historical version milestones and precise dates, it is recommended to refer to the official real-time page and complete the version nodes in subsequent iterations.

Technical advantages

Marker's technical orientation is "document parsing with layout understanding as the core" rather than simple text extraction.

Mechanism: Combined with layout detection OCR, dedicated processing of tables and formulas, and then reconstructing title level and block-level elements according to the document structure; complex scenes can be superimposed with LLM for semantic correction.

Effect: Compared with direct PDF text extraction, this structured reconstruction can better preserve the semantic level, making the recall quality of downstream retrieval and question answering higher.

Applicable scenarios: Most suitable for engineering scenarios that need to convert a large number of structured documents (papers, manuals, contracts, reports) into clean Markdown and feed them to RAG. For pure image scans or handwriting, the accuracy will drop.

How to use

Marker is mainly for developers and can be used as a command line tool or Python library, or you can use the Datalab cloud API instead:

• Installation: Install Python packages and dependencies according to the warehouse instructions (GPU configuration is recommended in some modes).
• Run Conversion: Convert single or batch PDFs to output Markdown/JSON.
• Adjust parameters on demand: Adjust OCR, table recognition and LLM enhancement options according to the document type.
• Integrated pipeline: Integrate the conversion results into the vectorization and retrieval process.
• Cloud alternative: If you don't want to build it yourself, call the Datalab API and use it as needed.

Implementation Tips: For first-time access, it is recommended to use a small batch of representative documents to evaluate the table and formula restoration rate, and then decide on local self-hosting or cloud API.

Product Pricing

Marker itself is open source and free, and can be used under the GitHub license; the supporting Datalab cloud API is a paid service.

Billing Dimension: There is zero licensing fee for open source use, and the cost comes from computing power and operation and maintenance; cloud APIs are usually billed based on the number of calls/number of pages. The specific price and quota are subject to real-time information on the Datalab official page.

Procurement Tip: If an enterprise uses a cloud API to process sensitive documents, it should confirm the data processing and retention strategy; if it is self-hosted, it should evaluate GPU resources and peak throughput capabilities.

Application scenarios

• RAG corpus preparation: Convert PDF knowledge base to clean Markdown to improve retrieval and Q&A quality.
• Document Digitization: Convert contracts, manuals, and reports into structured text in batches and store them in the database.
• Academic and technical content processing: Restore the formulas, tables and codes in the paper to facilitate secondary processing.

Key points for scenario verification: Each type of scenario should focus on checking the table restoration rate, multi-column layout sequence and formula accuracy, especially for structure-sensitive documents such as contracts and financial reports.

Applicable people

• AI/Data Engineer: A team that builds RAG and knowledge base and needs to batch structured documents.
• Research and Documentation Team: Researchers who need to convert a large number of papers or materials into editable text.
• Internal enterprise platform team: A team that handles sensitive documents in an isolated environment and requires that data not be sent out.

Not suitable for the boundary: Users who do not have R&D resources and only want to upload and convert via a graphical interface are more suitable to use the cloud API; when processing purely handwritten or low-quality scans, they need to accept a decrease in accuracy.

Summary and Outlook

Marker's core competitiveness lies in its open source, high-precision conversion of documents into structured text, and its ability to self-host processing of sensitive data. It has clear value to engineering teams building RAG and document automation pipelines. The community size of about 36k stars confirms its leading position in the document parsing track. The current limitations are: it requires certain engineering capabilities, there are still accuracy limits for scanned documents and complex forms, and the date of the unified external version has not been disclosed.

It is worth observing the accuracy of its LLM enhanced mode, new format support and the evolution of cloud API in the future. It is recommended that teams first use real document samples to evaluate table and formula restoration rates before choosing between self-hosted and cloud APIs; enterprises that handle sensitive documents should prioritize verifying the throughput and compliance boundaries of self-hosted solutions.

Marker version evolution

Marker continues to iterate on GitHub, and version evolution is reflected in the expansion of accuracy and format support.

Early stage

Version 0.x takes PDF to Markdown as the core and establishes the basic capabilities of high-precision layout restoration and table recognition.

Current stage

The 1.x mainline extends multi-format input JSON/HTML output and optional LLM enhancement mode on top of basic conversion, making the restoration quality of complex layouts (multiple columns, dense tables, formulas) more controllable. The specific patch number is subject to GitHub Releases, and there is no official announcement on the unified release date.

Comparison of competing products

Comparison Dimension	Marker	Competitor A	Competitor B
Core Differences	—	—	—
Price	—	—	—
Target Users	—	—	—

Note: The above comparison is based on product public information, and actual differences are based on user experience.

Version Info

• Marker 1.x ：The mainline version of GitHub's continuous iteration supports converting PDF and multi-format documents to Markdown/JSON/HTML, and integrates LLM enhancement mode to improve the accuracy of complex layouts. The specific patch number is subject to GitHub Releases, and there is no unified release date yet. （~2026-05）
• Marker 0.x ：The early version focused on converting PDF to Markdown, laying the foundation for high-precision layout restoration and table recognition. The official official release date has not been announced, and there is no official precise date yet. （~2024）