Wikipedia Database Dumps
Wikimedia Foundation
Every article and every past revision of every Wikipedia, free to download.
Full exports of all public Wikimedia wikis, including each Wikipedia language edition, as wikitext with page metadata, plus raw SQL tables for links, categories, and other structure. Full-history dumps include every revision, so you can study how articles were written and fought over, not just what they say now.
Access
- Cost
- FreeFree; downloads are rate limited to 3 connections per IP. Rendered HTML dumps come through the commercial Wikimedia Enterprise service.
- License
- Open licenseText under CC BY-SA 4.0 (and GFDL for older text); media files carry their own licenses
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadXML, SQL, OpenZim
What's in it
- One row is
- One record per page revision, with SQL side tables for links, categories, and redirects
- Covers
- All Wikipedia language editions (300+) and sister projects
- Years
- 2001-present
- Updated
- Monthly
- Size
- English articles-only dump is tens of GB compressed; full histories run to terabytes uncompressed
Quality
Complete, documented, and produced on a regular schedule by the platform itself; the canonical source for Wikipedia research.
- The older XML dump format is deprecated and slated to end in January 2027
- Wikitext needs parsing; templates do not expand on their own
- Article quality varies widely by language edition
Connects to
- Wikidata JSON Dumps: Each article links to a Wikidata item, which supplies structured facts about its subject
- SQuAD (Stanford Question Answering Dataset): SQuAD's passages are drawn from Wikipedia articles
- Common Crawl: Wikipedia is a small, high-quality slice usually mixed with web crawl text in LLM training
Questions it could answer
- Which articles have had the most reverts, and on what topics?
- How does article length on the same subject differ across language editions?
- How fast does an article grow after a major news event?