What is sparse-dense hybrid search in Pinecone?

Pinecone supports hybrid search by storing both a dense vector and a sparse vector per record, with the sparse vector typically generated by BM25 or a learned sparse model like SPLADE. Queries include both a dense and sparse vector plus an alpha weight to blend scores, returning results ranked by the combined score. Hybrid search improves recall on queries with rare terms, product names, or exact phrases that dense embeddings alone miss. Sparse indexes in Pinecone are stored alongside dense ones without separate infrastructure.