Qwen3 Reranker 0.6B
Qwen/Qwen3-Reranker-0.6B
published May 2025 · updated Apr 2026
Qwen3 Reranker 0.6B is a multilingual reranking model that reorders documents based on relevance to a query, supporting over 100 languages and a 32k context length.
specs
| Task | Text Reranking |
| Architecture | Dense Transformer (Qwen3) |
| Parameters | 0.6B |
| Context Length | 32k tokens |
| Supported Languages | 100+ languages |
| License | Apache 2.0 |
about this model
Qwen3-Reranker-0.6B is a multilingual text reranking model that reorders candidate documents by relevance to a given query, supporting over 100 languages and a context length of 32K tokens. It is part of the Qwen3 Embedding series, built on the Qwen3 dense foundation models and designed for text ranking tasks. The model is instruction-aware, allowing user-defined prompts to tailor behavior for specific tasks, languages, or domains. This capability typically yields a 1% to 5% improvement over default instructions.
Performance Benchmarks
Evaluated using top-100 candidates retrieved by Qwen3-Embedding-0.6B, the model achieves the following scores:
- MTEB-R: 65.80
- CMTEB-R (Chinese): 71.31
- MMTEB-R (Multilingual): 66.36
- MLDR: 67.28
- MTEB-Code: 73.42
- FollowIR: 5.41
Comparison with Other Rerankers
On the same evaluation setup, Qwen3-Reranker-0.6B outperforms comparable models:
| Model | MTEB-R | CMTEB-R | MMTEB-R | MLDR | MTEB-Code | FollowIR |
|---|---|---|---|---|---|---|
| Qwen3-Reranker-0.6B | 65.80 | 71.31 | 66.36 | 67.28 | 73.42 | 5.41 |
| Jina-multilingual-reranker-v2-base | 62.09 | 66.31 | 61.54 | 61.96 | 64.11 | 4.57 |
| gte-multilingual-reranker-base | 63.32 | 68.73 | 63.50 | 65.17 | 65.32 | 5.77 |
| BGE-reranker-v2-m3 | 63.53 | 65.56 | 62.34 | 61.25 | 65.08 | 3.64 |
Training and Availability
Training uses a multi-stage pipeline combining large-scale unsupervised pre-training, supervised fine-tuning on high-quality data, and model merging. The Qwen3 LLMs were also employed to synthesize training data. The model is released under the Apache 2.0 license. For further details, see the technical report (arXiv:2506.05176) and the blog post.
best for
- ·Reordering search results for multilingual web search
- ·Improving document retrieval in RAG pipelines across 100+ languages
- ·Code retrieval: ranking code snippets relevant to a natural language query
FAQ
It is best for reordering documents in multilingual retrieval tasks, including web search, RAG pipelines, and code retrieval, supporting over 100 languages.
With 0.6B parameters and 32k context, it is efficient and faster than larger 4B/8B variants, suitable for latency-sensitive applications.
It is released under the Apache 2.0 license.
Input is a query and a list of documents. Output is a relevance score (e.g., logit or probability) for each document, indicating relevance to the query.
Use the gigarouter OpenAI-compatible endpoint with your API key. Send a POST request to the rerank endpoint with the query and documents in the request body.
# rerank documents by relevance; billed per document curl https://gigarouter.ai/v1/rerank \ -H "Authorization: Bearer $GR_KEY" \ -d '{"model":"Qwen/Qwen3-Reranker-0.6B","query":"capital of France", "documents":["Paris is the capital of France.","Bananas are yellow."]}'
try it live
runs the real hosted model on a shared demo allowance · get your own key + $25 free →