AI Potluck
Contribute
Back to Gap Map Model components / Training & synthetic datasets

CulturaX

University of Oregon NLP (UONLP)
gated / Overall score: 3.0

Multilingual web corpus for pretraining containing 6.3 trillion tokens across 167 languages, combined and cleaned from mC4 and OSCAR by the University of Oregon NLP group. Its pipeline applies language identification, URL filtering, metric-based cleaning, document refinement and MinHash fuzzy deduplication, and it is distributed as roughly 16TB of parquet.

Openness

3 high confidence
3.0
license
follows mC4 + OSCAR-2301 terms
gated
auto(contact-info agreement)
card
present

Dataset card is rich but downloads require agreeing to share contact information (auto gate).

  • https://huggingface.co/datasets/uonlp/CulturaX recorded 2026-09-29

    Dataset page for uonlp/CulturaX states "The licence terms for CulturaX strictly follows those of" mC4 and OSCAR, and the gate reads "You need to agree to share your contact information to access this dataset".

Adoption

3 high confidence
3.0

Adoption is measured by Hugging Face downloads of the dataset over the trailing month, the only usage channel on record for CulturaX.

Capability

3 high confidence
3.0

CulturaX merges and cleans mC4 and OSCAR into one multilingual corpus, but the FineWeb-2 dataset card reports that FineWeb-2 outperforms it, along with CC-100, mC4 and HPLT, on FineTasks ablations across nine languages.

Verified 2026-09-27