BGE embeddings BAAI General Embedding
Chinese Beijing Academy of AI's open-weight embedding models — top of the MTEB benchmark, free to run yourself.
Models: bge-small/base/large-en, bge-m3 (multilingual + multi-function: dense + sparse + multi-vector), bge-reranker. M3 launched in January 2024 and quickly became the default for non-English RAG pipelines. Competition on open weights: nomic-embed, gte (Alibaba), Jina embeddings, mxbai-embed. Managed: OpenAI text-embedding-3, Cohere Embed v3, Voyage AI. MTEB (Massive Text Embedding Benchmark) is the leading evaluation leaderboard.