Common Crawl
Dataset
Web Corpus
The nonprofit web archive nearly every LLM ate first.
You're looking at a dataset record: a training-data collection — what's in it and which models learned from it.
Common Crawl has archived the open web monthly since 2008 — petabytes of pages that, filtered and deduplicated, form the backbone of most pre-training corpora (GPT-3's largest component; ancestor of C4, RefinedWeb, FineWeb). A tiny nonprofit's crawler quietly became the most economically consequential dataset on Earth — and the center of the consent-and-copyright fight.
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Used to train
Sources
Record reviewed as of 2026-01-15 — see Methodology.