AlibabaQwen Reranker
Qwen3-Reranker-4B
Qwen3 Reranker scores candidate relevance with task instructions. The published model supports over 100 languages and 32,000-token sequences.
Practical use
Practical guidance. Include near-miss documents in the evaluation: similar wording, wrong dates or the wrong entity. These cases show whether reranking improves relevance beyond simple topic similarity.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | Qwen/Qwen3-Reranker-4B |
| Output purpose | Relevance ranking; it does not create a stored document embedding.Qwen3-Reranker-4B model card |
| Licence | Apache 2.0.Qwen3-Reranker-4B model card |
Related models
- Qwen3-Reranker-8B
Alibaba · Qwen Reranker
Eight-billion-parameter multilingual reranker for instruction-guided relevance scoring.
Launch date not confirmedDownloadable - Qwen3-Reranker-0.6B
Alibaba · Qwen Reranker
A 600-million-parameter reranker that scores query-document relevance across more than 100 languages.
Launch date not confirmedDownloadable - Qwen3.7 Text Rerank
Alibaba · Qwen Reranker
Hosted text reranker with a limit of 30,000 tokens per candidate.
Launch date not confirmedAPI