Qwen3-Embedding 4B is a 4-billion-parameter multilingual embedding model built on the Qwen3 foundation, designed specifically for text embedding and ranking tasks. With 36 transformer layers, the model produces high-quality vector representations for text and code across retrieval, classification, clustering, and bitext mining workflows. The architecture supports instruction-conditioned embeddings, allowing developers to guide representation quality for specific languages, domains, or tasks. Its design reflects the Qwen series' emphasis on cross-lingual representation learning, enabling semantic understanding across more than 100 languages and various programming languages.
The Qwen3 Embedding series builds upon the dense foundational models of the Qwen3 family, inheriting their multilingual capabilities, long-text understanding, and reasoning skills. The 8B variant in particular achieved state-of-the-art performance, ranking No. 1 on the MTEB multilingual leaderboard with a score of 70.58 as of June 2025. The model is available as open weights under Apache 2.0, making it suitable for developers who need to deploy customized retrieval pipelines or combine embedding with reranking modules. This combination of strong benchmark results, flexible dimensionality output options, and instruction-tuning support positions the 4B variant as a practical choice for developers balancing efficiency with multilingual retrieval quality.
