Qwen3 Embedding and Reranker
Open Apache 2.0 text embedding and reranking models (0.6B, 4B, 8B) from Qwen3; the 8B ranked first on MTEB multilingual with 70.58 on 2025-06-05.
Embedding and reranker models initialised from Qwen3 rather than a separate encoder, with 32K context and instruction-aware vectors. Qwen3-VL-Embedding and Qwen3-VL-Reranker followed in January 2026.
- Date
- Thursday, 5 June 2025
- Lab
- Alibaba (Qwen)
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| MTEB multilingual score (8B) | 70.58 Ranked No. 1 as of 2025-06-05 per Qwen; 8B uses 4096-dim embeddings, 32K context | company |
| Reranker MTEB-R (4B) | 69.76 Company table; 8B reranker 69.02 | company |
Leaderboard rank is a company statement on a public leaderboard that has since moved. Blog dated 2025-06-05.
Sources
- qwenlm.github.io/blog/qwen3-embedding/
- huggingface.co/api/models?author=Qwen&sort=createdAt&direction=-1&limit=60&skip=60
- arxiv.org/abs/2506.05176
This record was checked against its sources on 6 October 2026. How we check