Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models

By Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin and 2 more. arXiv.org, 2025

1,074 citations recorded by the source corpus.

In this work, we introduce the Qwen3 Embedding series, a significant advancement over its predecessor, the GTE-Qwen series, in text embedding and reranking capabilities, built upon the Qwen3 foundation models. Leveraging the Qwen3 LLMs' robust capabilities in multilingual text understanding and generation, our innovative multi-stage training pipeline combines large-scale unsupervised pre-training with supervised fine-tuning on high-quality datasets. Effective model merging strategies further ensure the robustness and adaptability of the Qwen3 Embedding series. During the training process, the Qwen3 LLMs serve not only as backbone models but also play a crucial role in synthesizing high-quality, rich, and diverse training data across multiple domains and languages, thus enhancing the training pipeline. The Qwen3 Embedding series offers a spectrum of model sizes (0.6B, 4B, 8B) for both embedding and reranking tasks, addressing diverse deployment scenarios where users can optimize for either efficiency or effectiveness. Empirical evaluations demonstrate that the Qwen3 Embedding series achieves state-of-the-art results across diverse benchmarks. Notably, it excels on the multilingual evaluation benchmark MTEB for text embedding, as well as in various retrieval tasks, including code retrieval, cross-lingual retrieval and multilingual retrieval. To facilitate reproducibility and promote community-driven research and development, the Qwen3 Embedding models are publicly available under the Apache 2.0 license.

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/41a8c615b8fefdddbb28e1c0a9edcaa446d15451

Citations in the corpus

• Bridging Language and Items for Retrieval and Recommendation: Benchmarking LLMs as Semantic Encoders (2024, neutral): “Among these, we further benchmark Qwen3-Embedding models of varying sizes (0.6B, 4B, and 8B) (Zhang et al., 2025) to better understand the scaling behavior of LLM-based semantic encoders.”
• Mixture-of-Experts Language Models Can Be Strong and Efficient Retrievers (2026, unknown): “…recipe for adapting decoder-only LLMs into bi-encoders (Ma et al., 2024), and subsequent work explored LLM-generated training data (Wang et al., 2024), bidirectional attention (BehnamGhader et al., 2024), latent-attention pooling (Lee et al., 2025a), and multi-stage training (Zhang et al., 2025).”
• OntoLearner: A Modular Python Library for Ontology Learning with Large Language Models (2026, unknown): “We also include Qwen3 Embeddings (Zhang et al. 2025), a family of proprietary multilingual embedding models available in 0.6B, 4B, and 8B parameter sizes (Qwen3-0.”
• ALEE: Any-Language Evaluation of Embeddings via English-Centric Minimal Pairs (2026, unknown): “We evaluate the Qwen3-Embedding family (0.6B, 4B, 8B) (Zhang et al., 2025) under the same TACC protocol, comparing against some of the sub-600M encoder models (Table 3).”
• Zero-Shot Semantic Re-Identification for Autonomous Driving: A VLM Baseline Study (2026, unknown): “…[15] • Gemma 4 E2B [16] We pair these with four text embedding models: • Embedding Gemma 300M [17] • Nomic Embed v2 305M [18] • Qwen3-Embedding 4B [19] • Qwen3-Embedding 8B [19] This yields a cross-product of description generators and encoders, allowing us to isolate the impact of semantic…”
• IdioLink: Retrieving Meaning Beyond Words Across Idiomatic and Literal Expressions (2026, unknown): “5 (Qwen et al., 2025), the Gemma-based BGE model (Chen et al., 2024), and the Qwen3 Embedding models (Zhang et al., 2025c).”
• The TellTail of Embeddings: Fingerprinting Retrievers in Black-Box Systems (2026, unknown): “6b and qwen-4b [41]; e5-small , e5-base , and e5-large [42]; gte-modernbert , gte-large-v1.”
• Towards Retrieval Augmented Generation in High-Energy and Astroparticle Physics (2026, unknown): “We will also compare the performance of Specter2 with two general purpose embedders Qwen3 [19] and Nomic [20].”
• A Systematic Multi-Domain Evaluation of Document Retrievers (2026, unknown): “Retrievers. docT5query [58] expands documents by generating multiple queries for each and retrieving with BM25. query2doc [80] instead expands the query, generating a pseudo-document and again retrieving with BM25, while HyDE [19] generates a hypothetical document and retrieves with the Contriever.”
• Q-TIE: A Lightweight and Generalizable Re-ranking Framework for Temporal Information Retrieval (2026, unknown): “Compared to general neural re-rankers (Zhang et al., 2025) and prompt-based LLM re-rankers (Chen et al., 2026), Q-TIE provides temporal specialization at substantially lower cost (0.6B parameters), making it a practical drop-in to achieve temporal awareness for modern RAG pipelines that often incur…”
• Q2D-Web: A Large-Scale Benchmark for Retrieval in Agentic RAG Systems (2026, unknown): “6B (Zhang et al., 2025b) only models released before 1 January 2025 contribute to subcorpus construction, and ℳ spans lexical, dense, and late-interaction retrievers (Appendix A), so the held-out models are tested against a pool built from all three families rather than from a single one.”
• REIGN: Refurbished Embeddings with Integrated Guidance Networks for Efficient Context-Length Scaling (2026, unknown): “A parallel line instruction-tunes billion-scale LLMs into embedding models (Lee et al., 2025; Choi et al., 2024; Zhang et al., 2025; Meng et al., 2024); these are accurate but computationally heavy, and beyond most groups’ training budgets even with parameter-efficient adapters or quantisation.”
• Giga-Embeddings: Mixture-of-Experts Encoders for High-Throughput Text Embeddings (2026, unknown): “The 480M and 3B models are dense bidirectional Qwen3 encoders (Zhang et al., 2025), while the largest model is a bidirectional DeepSeekMoE-style encoder (Dai et al., 2024) with 10B total parameters, 64 routed experts, one shared expert, and top-4 routing.”