Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models
- Source
- qwenlm.github.io
- Author
- Qwen Team
- Date
Context
Qwen's Qwen3 family separates finding plausible search results from judging which results best answer a query. Its embedding models turn individual texts into numerical representations that support retrieval. Its models then read a query and candidate document together and assign a relevance score. Developers can combine these components into a search system. The release provides both types in several sizes under Apache 2.0.
The reported reranking reflects that two-stage design: every reranker received the same top 100 candidates retrieved by the smallest embedding model. The scores therefore measure how well each model reorders that shortlist. They help assess the second stage without establishing that a reranker can recover relevant documents absent from the initial results.
- embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
- reranking — A second pass that re-scores retrieved candidates by reading each one against the query, fixing the ordering that fast vector search got approximately right.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
“We release Qwen3 Embedding series, a new proprietary model of the Qwen model family.”
Checking sign-in…
Loading comments…