Overview
The Qwen3 Reranker models score how well a document answers a query, the second stage of most retrieval pipelines after a fast embedding search. They share the Qwen3 foundation with the Qwen3 Embedding models and accept task instructions to adapt ranking to a use case.
Highlights
- 0.6B, 4B and 8B sizes
- Over 100 languages, including programming languages
- Instruction-aware: a task instruction can be added to the input
- 32K-token context
Typical uses
Reranking search results in RAG systems, enterprise search and code retrieval.
Things to consider
- Rerankers are slower than embeddings because every query-document pair is scored; apply them to a short candidate list.
- The model card notes that instructions are best written in English, even for multilingual data.
Specifications are taken from the model card on Hugging Face and the provider’s announcement. Benchmark results depend on evaluation settings; see how AI benchmarks work.