Guide · Search & RAG

Jev as a reranker: an honest negative

Jason Zhu tested Jev reranking on 164 real queries.

Open x.com ↗

拿 Jev 做搜索重排，我先泼一盆冷水：单独用，它没打赢向量检索

TypeSafe 的 Jev 这阵子很火，一堆项目拿它做重排。我们在 Agent Skills Hub 的 33,047 条目录上认真测了一次，164 条中英文真实查询，9,831 对分级标注，整套只花了 2.6 美元

三个结论

01｜单独重排，约等于没赢

Jev 重排 bge-m3

Sep 18, 2026Show the live post from X

How builders describe it

Jason Zhu tested Jev reranking on 164 real queries. Alone it did not clearly beat vector search; fused with it, it did. In Chinese.

A measured search reranking run over 33,047 catalog entries, 164 real queries, and 9,831 labelled pairs, reporting how Jev compares with BM25 and bge-m3.

The decision Jev makes

Rank candidates against a query, or answer whether a passage is relevant.

Where it fits

Search pipelines retrieve too much. These projects use Jev as a reranker or relevance check after the first search, so only useful results reach the user or the model. All 108 search & rag projects →

Related projects