[KongchangAI]
Concept

Context Focus

A context compression feature introduced by Cohere in Rerank 5 that performs relevance judgment and passage-level extraction together during reranking, retaining only the passages of a document relevant to the query, with an adjustable retention amount to reduce token costs.

Timeline (last 90 days)

Oct 22

The core mechanism of Rerank 5 is Context Focus: it first locates the passages in a document relevant to the query, then scores relevance based on those passages, rather than compressing and scoring the entire text.

Unverified50%
Oct 21

Rerank 5 introduces a new feature called Context Focus, which can extract only the specific passages in a document relevant to the query to reduce token consumption, and it is optional to enable with an adjustable amount of retained context.

Unverified50%
Oct 21

Context Focus merges relevance scoring and passage-level extraction into the reranking stage, outputting the specific passages matched in the document rather than the entire document.

Unverified50%

All Facts (3)

Source Articles