• Elasticsearch 9.0 & 8.18 released, available on Elastic Cloud and self‑managed. • Better Binary Quantization (BBQ) now GA, 5x faster than OpenSearch BBQ. • BBQ offers 20% higher recall, 8x-30x faster throughput with SIMD. • Elastic first vector database vendor to implement BBQ, boosting real‑world search. • BBQ outperforms OpenSearch FAISS: 5x faster queries, 3.9x higher throughput, same accuracy. • Simplified API for quantized vector rescoring improves developer experience.

Article Summaries:

  • Elastic released Elasticsearch 9.0 and 8.18, adding the new Better Binary Quantization (BBQ) vector‑search engine. BBQ, now generally available, delivers up to 20 % higher recall and 8‑30× faster throughput than previous methods, and is reported to be 5× faster than OpenSearch’s FAISS implementation while maintaining the same accuracy. The update also simplifies the quantized rescoring API, supports multi‑stage models such as ColPali and ColBERT, and introduces out‑of‑the‑box semantic search with ELSER and e5 models. Elastic’s open inference API lets users plug in third‑party embeddings and reranking services, broadening developer flexibility.

Sources: