Developer: BAAI · Open-source embedding model
A multilingual embedding model supporting dense, sparse and multi-vector retrieval for short text and long documents.
No official API pricing published
Open-source model weights can be downloaded and self-hosted under their license. Hosting, compute and third-party API inference may incur separate charges. No verified, directly purchasable API quote is listed here.
- Model ID
- BAAI/bge-m3
- Category
- Embedding
- Version
- m3
- Languages
- multilingual
- Input length
- 8,192 tokens
- Embedding dimensions
- 1024
- License
- MIT
- Official announcement
- 2024-01-30
- Verified
- 2026-10-15
Maximum input sequence length in tokens.
Use cases: RAG, semantic search, knowledge base, retrieval.
Retrieval methods: Dense retrieval, Sparse lexical retrieval, Multi-vector retrieval.
Official sources and model revision
https://huggingface.co/BAAI/bge-m3
https://huggingface.co/api/models/BAAI/bge-m3
https://huggingface.co/BAAI/bge-m3/raw/5617a9f61b028005a4858fdac845db406aefb181/README.md
https://huggingface.co/BAAI/bge-m3/raw/5617a9f61b028005a4858fdac845db406aefb181/config.json
https://huggingface.co/BAAI/bge-m3/raw/5617a9f61b028005a4858fdac845db406aefb181/tokenizer_config.json
https://github.com/FlagOpen/FlagEmbedding
Revision: 5617a9f61b028005a4858fdac845db406aefb181