Alibaba Unveils Qwen3-Embedding and Qwen3-Reranker for Multilingual Text Processing

Alibaba Group has launched the Qwen3-Embedding and Qwen3-Reranker series, setting new standards in multilingual text embedding and relevance ranking. These models, built on the Qwen3 foundation, are available in 0.6B, 4B, and 8B versions and support 119 languages. They are open-sourced under the Apache 2.0 license and accessible via Alibaba Cloud APIs.

The Qwen3-Embedding series excels in text embedding, retrieval, and reranking tasks, achieving state-of-the-art performance across multiple benchmarks. The 8B size embedding model ranks first on the MTEB multilingual leaderboard with a score of 70.58. The reranking models significantly improve search relevance, making them ideal for various applications.

These models are designed using dual-encoder and cross-encoder architectures, enhancing multilingual, cross-lingual, and code retrieval capabilities. The training framework follows a multi-stage paradigm, improving training efficiency and deployment performance across different scenarios. Alibaba plans to further optimize the Qwen foundation model to enhance these capabilities.