Multilingual E5 Text Embeddings: A Technical Report

By Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, Furu Wei. arXiv.org, 2024

521 citations recorded by the source corpus.

This technical report presents the training methodology and evaluation results of the open-source multilingual E5 text embedding models, released in mid-2023. Three embedding models of different sizes (small / base / large) are provided, offering a balance between the inference efficiency and embedding quality. The training procedure adheres to the English E5 model recipe, involving contrastive pre-training on 1 billion multilingual text pairs, followed by fine-tuning on a combination of labeled datasets. Additionally, we introduce a new instruction-tuned embedding model, whose performance is on par with state-of-the-art, English-only models of similar sizes. Information regarding the model release can be found at https://github.com/microsoft/unilm/tree/master/e5 .

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/85ceb93a797481b5203b36f4e77a0828107c42fd

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. A. Engel
2. Tony Chiang
3. Anand D. Sarwate
4. Reilly Cannon
5. Max Vargas
6. Hong-Liu Cao

Citations in the corpus

• IdioLink: Retrieving Meaning Beyond Words Across Idiomatic and Literal Expressions (2026, unknown): “We assess encoder models: Contriever-based TART (Asai et al., 2023), the instruction-tuned multilingual E5 (Wang et al., 2024c), and Nomic-Embed-v2 (Nussbaum and Duderstadt, 2025).”
• What Drives Cross-lingual Ranking? Retrieval Approaches with Multilingual Language Models (2025, unknown): “Retrieval-oriented encoders improve alignment using ranking objectives and supervision from parallel or translation-based data, as in LaBSE [39], LASER [40], and multilingual-E5 [41], achieving strong results on multilingual benchmarks such as MMTEB [42].”
• Evaluating Embedding Generalization: How LLMs, LoRA, and SLERP Shape Representational Geometry (2025, unknown): “More recent families of embedding models, such as E5 [17] and GTE [18], scale contrastive objectives while incorporating multilingual and task-adapted training.”
• TurkEmbed: Turkish Embedding Model on Natural Language Inference & Sentence Text Similarity Tasks (2025, unknown): “In addition, the exploration of multilingual models such as XLM-ROBERTa [23] and multilingual E5 [24] has opened the doors for the learning of cross-lingual transfer in Turkish NLP.”
• Chunk Twice, Embed Once: A Systematic Study of Segmentation and Representation Trade-offs in Chemistry-Aware Retrieval-Augmented Generation (2025, unknown): “The model set also includes Nomic Embed [61] and multilingual E5 and BGE variants [59, 62].”
• Mixture of Languages: Improved Multilingual Encoders Through Language Grouping (2025, unknown): “... ing MEXMA (Janeiro et al., 2025), mE5 (Wang et al., 2024), and Jina-Embeddings-V3 (Sturua et al., 2024).”
• Q2D-Web: A Large-Scale Benchmark for Retrieval in Agentic RAG Systems (2026, unknown): “…use BM25 (Robertson and Zaragoza, 2009), ColBERTv2 (Santhanam et al., 2022), the dense models arctic-l-v2 (Yu et al., 2024), bge-m3 (Chen et al., 2024a), e5-large-instruct (Wang et al., 2024), gte-multilingual-base (Zhang et al., 2024), mxbai-large-v1 (Lee et al., 2024; Li and Li, 2023), nomic-v1.”
• Choosing a Text Embedding Model: A Practical Benchmarking and Decision Framework (2026, unknown): “Multilingual-E5-large (mE5-L) [9] Open-source Yes https://huggingface.co/intfloat/ multilingual-e5-large LaBSE [12] Open-source Yes https://huggingface. co/ sentence-transformers/LaBSE Paraphrase- Multilingual-MPNet (mMP-Net) [14, 13] Open-source Yes https://huggingface. co/ sentence-transformers /…”
• HAKARI-Bench: A Lightweight Benchmark for Comparing Retrieval Architectures and Efficiency Settings under Unified Conditions (2026, unknown): “For example, Multilingual E5 (Wang et al., 2024a) is reported to include MS MARCO and MIRACL in its supervised fine-tuning data mixture, so the NanoMIRACL and NanoBEIR-en numbers may be affected.”
• Granite Embedding Multilingual R2 Models (2026, unknown): “Encoder-based embedding models (Wang et al., 2022; Xiao et al., 2023; Chen et al., 2024; Merrick et al., 2024; Zhang et al., 2024; Wang et al., 2024a; Nussbaum et al., 2024; Awasthy et al., 2025b) are widely used for retrieval due to their low inference latency and small memory footprint compared…”
• MLAIRE: Multilingual Language-Aware Information Retrieval Evaluation Protocal (2026, unknown): “By contrast, multilingual-e5 follows multi-monolingual contrastive training, where query–positive pairs are constructed within the same language [32].”
• Beyond Chunk-Then-Embed: A Comprehensive Taxonomy and Evaluation of Document Chunking Strategies for Information Retrieval (2026, unknown): “• multilingual-e5-large (E5-large) [Wang et al.(2024)]: A 560M parameter multilingual model that serves as a standard baseline in retrieval research.”
• GigaEmbeddings: Efficient Russian Language Embedding Model (2025, unknown): “We compare two ways of prompting model: prefix (Nuss-baum et al., 2024, Wang et al., 2024c) and instruct (BehnamGhader et al., 2024, Lee et al., 2025).”
• ChEmbed: Enhancing Chemical Literature Search Through Domain-Specific Text Embeddings (2025, unknown): “Text Embedding Models and Domain Adaptation Text embedding models transform texts into fixed-length vectors capturing semantic meaning, enabling effective retrieval in tasks such as retrieval-augmented generation (RAG) [11].”