LightOn-rerank blog · model collection
0.8B + 2B + 4B · text or page images · listwise generation

MULTIMODAL LISTWISE RERANKING PLAYGROUND

Let four candidates compete

Give the model a query and four text passages or four document pages. It compares them together and returns an order from most to least relevant.

Model size

2B default · 0.8B fastest · 4B strongest visual result

Candidate format

Four candidates

Candidate numbers identify the input order used in the generated permutation.

Model ranking

A generated relative order, not four independent relevance scores.