Elasticsearch tops specialized vector databases for RAG

The "vector-only" hype is hitting a reality check as developers realize that semantic similarity often fails where precise keyword matching excels. Hybrid search combining BM25 and vector embeddings is the current gold standard for RAG, yet many new AI devs ignore the 15+ years of optimization in Lucene-based engines. Elasticsearch and OpenSearch offer mature enterprise features like complex filtering, aggregations, and battle-tested scaling that new vector-native DBs are still trying to build. Small BERT models can be hosted directly on CPUs within these clusters, making them highly viable for low-cost, local-first RAG without requiring massive GPU infrastructure.