BAAI:BGE-M3

The BGE-M3 model is a multi-functional, multi-lingual, and multi-granularity text embedding framework that integrates three retrieval methods: dense retrieval, multi-vector retrieval, and sparse retrieval. Capable of processing over 100 languages and inputs ranging from short sentences to lengthy documents (up to 8,192 tokens), it delivers state-of-the-art performance in multilingual and cross-lingual tasks, topping benchmarks like MIRACL and MKQA. Additionally, BGE-M3 excels in long-document retrieval, achieving strong results on datasets such as MLDR and NarrativeQA, solidifying its versatility across diverse text granularities and applications.