Embedding Models for GraphRAG: Selection and Benchmarks

Embedding models: text-embedding-3-small (1536 dim, good balance), text-embedding-3-large (3072 dim, highest quality), all-MiniLM-L6-v2 (384 dim, open-source, local), BGE-large-en-v1.5 (1024 dim, competitive open-source). Higher dimensions improve quality but increase costs. Normalize embeddings for cosine similarity. Consider dimension reduction for large-scale deployments.