BGE-M3 for Multilingual and Hybrid Retrieval
BGE-M3 appears in our text embedding API comparison for 100+ language support and dense + sparse + multi-vector retrieval from a single model.
BGE-M3 — Multilingual Embedding Model for Search and Retrieval 🔢📝❓ → 📝
Performance
53.4sTypical run time
~173sCold start (first call)
293Total runs
Data sampled 2026-07-12
BGE-M3 by BAAI is one of the most versatile multilingual embedding models available. Run it on Replicate to generate embeddings for search, clustering, or classification across languages.
About
BGE-M3 is a multilingual embedding model that supports over 100 languages. It generates dense vector representations from text, making it useful for semantic search, document retrieval, and RAG (retrieval-augmented generation) pipelines.
The model supports three retrieval modes — dense, sparse, and multi-vector — so you can pick the approach that fits your accuracy and speed requirements.
Why BGE-M3 stands out
- Works across 100+ languages without separate models per language
- Handles long documents up to 8,192 tokens
- Supports dense, sparse, and ColBERT-style multi-vector retrieval
- Strong performance on multilingual benchmarks like MIRACL and MTEB
Example Output
Output
[[0.626 0.3477]
[0.3499 0.678 ]]
[0.3499 0.678 ]]
Performance Metrics
53.45s
Prediction Time
172.98s
Total Time
All Input Parameters
{
"max_length": 4096,
"sentences_1": "What is BGE M3?\nDefination of BM25",
"sentences_2": "BGE M3 is an embedding model supporting dense retrieval, lexical matching and multi-vector interaction.\nBM25 is a bag-of-words retrieval function that ranks a set of documents based on the query terms appearing in each document",
"embedding_type": "dense"
}
Input Parameters
- max_length
- Maximum length of the input for dense embeddings, use a smaller value to speed up the encoding process
- sentences_1 (required)
- Input Sentence list 1 - Each sentence should be split by a newline
- sentences_2 (required)
- Input Sentence list 2 - Each sentence should be split by a newline
- embedding_type
- Type of embedding to use
Output Schema
Output
Example Execution Logs
Sentences_1 split out: ['What is BGE M3?', 'Defination of BM25'] Sentences_2 split out: ['BGE M3 is an embedding model supporting dense retrieval, lexical matching and multi-vector interaction.', 'BM25 is a bag-of-words retrieval function that ranks a set of documents based on the query terms appearing in each document'] encoding: 0%| | 0/1 [00:00<?, ?it/s] encoding: 100%|██████████| 1/1 [00:23<00:00, 23.31s/it] encoding: 100%|██████████| 1/1 [00:28<00:00, 28.09s/it] encoding: 0%| | 0/1 [00:00<?, ?it/s] encoding: 100%|██████████| 1/1 [00:22<00:00, 22.75s/it] encoding: 100%|██████████| 1/1 [00:25<00:00, 25.20s/it]
Version Details
- Version ID
3af6c861256a2a8e07a54a478813e6632f339f05235b59374f292f4759555bfb- Version Created
- February 7, 2024