Qwen 3.5 delivers production-grade embedding and reranking models that combine multilingual support, strong reasoning capabilities, and compact sizes for efficient inference and cost management.
The Qwen3 Embedding models (0.6B, 4B, 8B) rank #1 on MTEB's multilingual leaderboard, handling 100+ languages with consistent quality. The 8B model achieves a 70.58 MTEB score, matching or exceeding much larger competitors. Matryoshka Representation Learning enables variable embedding dimensions, letting you optimize storage and query latency for your use case. The complementary Qwen3-Reranker provides cross-encoder scoring for re-ranking dense retrieval results, significantly improving search relevance.
Zilliz Cloud integrates seamlessly with Qwen3 embeddings through our managed vector database service. You can vectorize multilingual documents, store them in Zilliz Cloud, and apply Qwen3-Reranker for two-stage retrieval—all without managing infrastructure. This architecture delivers enterprise-grade search quality while simplifying deployment and scaling.