The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale
Summary
FineWeb is a 15-trillion token pretraining dataset derived from 96 Common Crawl snapshots, which the authors say produces better-performing LLMs than other open pretraining datasets. They document and ablate design choices, including deduplication and filtering strategies. They also introduce FineWeb-Edu, a 1.3-trillion token collection of educational text filtered from FineWeb, and report that LLMs pretrained on FineWeb-Edu perform dramatically better on knowledge- and reasoning-intensive benchmarks such as MMLU and ARC. Along with the datasets, they publicly release their data-curation codebase and the models trained during ablation experiments. [S1]
Claims
- id: C1 claim: 'FineWeb is a 15-trillion token dataset derived from 96 Common Crawl snapshots.' source: S1 evidence: "a 15-trillion token dataset derived from 96 Common Crawl snapshots"
- id: C2 claim: 'FineWeb-Edu is a 1.3-trillion token collection of educational text filtered from FineWeb.' source: S1 evidence: "FineWeb-Edu, a 1.3-trillion token collection of educational text filtered from FineWeb"
- id: C3 claim: 'The authors publicly release the data-curation codebase and ablation-trained models.' source: S1 evidence: "we publicly release our data curation codebase and all of the models"
Verification
- Code available: Paper: https://huggingface.co/datasets/HuggingFaceFW/fineweb and https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu (both HTTP 200 on 2026-11-12). https://github.com/huggingface/datatrove also returned HTTP 200. Not run.
- Reproduced: not reproduced (2026-11-12). Code not run.
- Related work: This batch also has a verification note for The Pile. Live note 2302-13971-llama.
- Limits: Claims that FineWeb beats other open pretraining datasets, and that FineWeb-Edu helps on MMLU and ARC, are author-reported (v2, 31 Oct 2024).
- License: CC BY 4.0 on the abs page (2026-11-12; icon by-4.0.png; http://creativecommons.org/licenses/by/4.0/). Submitted 25 Jun 2024 (v1); last revised 31 Oct 2024 (v2). Sources: [S1] https://arxiv.org/abs/2406.17557, v2 (31 Oct 2024). checked_on: 2026-11-12.
When to apply
Use when comparing open web pretraining corpora, or when you need the FineWeb and FineWeb-Edu curation recipe.
Facts
- id: F1 claim: 'FineWeb is a 15-trillion token dataset derived from 96 Common Crawl snapshots.' confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
- id: F2 claim: 'FineWeb-Edu is a 1.3-trillion token educational subset filtered from FineWeb.' confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
- id: F3 claim: 'The authors publicly release the datasets, a data-curation codebase, and models from their ablations.' confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
Sources
- [S1] arXiv 2406.17557, https://arxiv.org/abs/2406.17557. License: CC BY 4.0. v2 (31 Oct 2024). Accessed 2026-11-12.
License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted). Paper is CC BY 4.0.
License: CC-BY-4.0 · Summary by iinventedlight (AI-assisted). Source: Guilherme Penedo, Hynek Kydlíček, Loubna Ben allal, Anton Lozhkov, Margaret Mitchell, Colin Raffel, Leandro Von Werra, Thomas Wolf, The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale (CC BY 4.0).
Cite this
'iinventedlight (AI-assisted)'. 2026. "The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale." iinventedlight Library, iil:note:sx48pt33, version of 2026-11-12. https://iinventedlight.com/library/notes/2406-17557-fineweb/. License: CC-BY-4.0.
@misc{iil_note_sx48pt33,
author = {{'iinventedlight (AI-assisted)'}},
title = {{The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale}},
howpublished = {iinventedlight Library},
year = {2026},
date = {2026-11-12},
url = {https://iinventedlight.com/library/notes/2406-17557-fineweb/},
note = {Permanent ID iil:note:sx48pt33. License: CC-BY-4.0.},
}
Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.
For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2406-17557-fineweb.md.