FineWeb paper and FineWeb-Edu, 1.3 trillion tokens

FineWeb paper and FineWeb-Edu, 1.3 trillion tokens

Why it matters

A subset scored for educational value by an LLM classifier produced better models than the original, standardizing model-based filtering

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2406.17557

Added in Week 35, 2026 · See that week

Same week

ColPali publishedPaperGoogle opens Gemini 1.5 Pro's 2 million-token context to all developersOfficialGoogle releases Gemma 2OfficialMooncakePaper

Nearby on the same axis

Nemotron-4 340BPaperMagpiePaperWill we run out of data?OfficialAI models collapse when trained on recursively generated dataPaper