The Stack v3 ships 4.9 trillion tokens across 713 languages, and this time the file contents are actually in the dataset.

Two artifacts: a training subset of 15.9 TB / ~4.9T tokens from 173M repositories, and a full pre-filter corpus of 113.7 TB across 770 languages from 224M repositories, crawl cutoff August 7 2025, ODC-By v1.0 with source files keeping original licenses (Hugging Face). The scale jump over v2 is roughly 9x in training tokens (550B → 4.9T), but the operationally important change is that file contents are embedded inline rather than referenced by file ID. Self-contained instead of requiring a separate fetch. Dedup moved to language-agnostic near-dedup with Jaccard verification after connected-component clustering, rows grouped one-per-repository for full-repo-context pretraining, opt-outs removed before each patch release. 182,428 downloads last month.