Jina AI presents Late Chunking
Late Chunking presented — the whole document is first passed through a long-context embedding model, then the token embeddings are pooled chunk by chunk
Why it matters
Solved the problem of chunking cutting off context by changing the order of embedding
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2409.04701Added in Week 35, 2026 · See that week