Dense vs multi-vector in one line: one vector per text vs one per token

tomaarsen · x · 2026-08-18

A dense model compresses a whole text into one vector; a multi-vector model keeps one vector per token and scores query against document with MaxSim — for each query token, take its best match in the document, then sum. Nothing gets averaged away. This is part of a thread accompanying a blog post by Tom Aarsen, Antoine Chaffin and Raphael Sourty covering checkpoint loading, scoring, search stacks, page-image retrieval and keeping the index affordable.

Related event: Sentence Transformers v6.0 ships with first-class late interaction models(33 posts)→

Original post →

More from coding & agent

coding & agent channel →