E5 is a family of open text embedding models that turn text into vectors for search, clustering and classification, trained on a large set of text pairs.
In plain words
To teach a model which texts mean similar things, E5 learns from a big curated collection of text pairs rather than hand-labelled data. This is called contrastive training with weak supervision. Without any labelled data, it was the first model to beat the classic BM25 keyword method on the BEIR search benchmark. When using E5, you add a short prefix like "query:" or "passage:" to each text.
A quick example
A typical search setup: store every document as "passage: " plus its text, embedded once. When someone searches, embed "query: " plus their question and return the passages whose vectors are closest.
Where you'll meet it
- Retrieval for RAG systems and search engines.
- Clustering and classifying text.
Related
Sources
- PaperText Embeddings by Weakly-Supervised Contrastive Pre-training, Wang et al. (arXiv) · read 29 Sept 2026
- Docsintfloat/e5-large-v2, Hugging Face · read 29 Sept 2026
- DocsE5 Text Embeddings, Microsoft unilm (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.