The Stack v3 expands to 114 TB with 224M repos and fixes a dedup bug
anton_lozhkov · x · 2026-07-23
The Stack v3 is out as the largest open code dataset ever released: 114 TB, 224M repositories, 770 languages, and roughly 5T deduplicated, filtered source-code tokens.
Key points:
- The v2 pipeline had a bug: MinHash shingles were tokenized with \W instead of \W+, causing many empty shingles and excessive hash collisions.
- That bug made v2 over-deduplicate heavily.
- v3 fixes the regex, runs MinHash-LSH across all languages together, and validates every candidate pair against its Jaccard estimate.
- The release claims zero false positives.
- It is fully open and excludes restrictively licensed code.
The accompanying chart shows large token growth across major languages, with C++ reaching 802B tokens and the all-language total at about 4.9T.
Related event: The Stack v3 releases 5T code tokens across 700+ programming languages(2 posts)→
More from Infra
- Chinese open-source models may be reinforcing NVIDIA’s inference moat — zephyr_z9 · 2026-07-23
- AI portability is really about identity, traces, and rollback — not model IDs — Early-Matter-8123 · 2026-07-23
- OpenShip challenges AWS and Vercel with open-source deploys, MCP, and self-hosting — eyishazyer · 2026-07-23
- Google’s AI push looks increasingly centered on infrastructure — Kindly_Ganache9027 · 2026-07-23
- COLMAP merges DEGENSAC into main RANSAC, boosting plane-heavy scenes — ducha_aiki · 2026-07-23
- xyOps says it will auto-file incident tickets with full context when jobs fail — tom_doerr · 2026-07-23