Which vector database handles metadata filtering best?

Qdrant has the most expressive filtering system, with nested conditions, geo filters, full-text match, and payload indexing that makes filtered search scale linearly with filter selectivity rather than collection size. Weaviate and Milvus both support filtered search with good performance when metadata fields are indexed. Pinecone supports metadata filtering but with simpler operators and historically slower filtered-vector performance on highly selective filters. pgvector inherits PostgreSQL's full SQL filtering, which is the most flexible but depends on how well the planner combines vector and WHERE clauses.