TL;DR
Quiet week for retrieval. One notable infra update: Elastic introduced a Vector Search–optimized hardware profile on Microsoft Azure, making it easier to deploy cost-efficient ANN/quantized search stacks at scale. No other material late‑interaction, learned‑sparse, or ANN‑infrastructure developments surfaced since October 15, 2025.
Elastic adds “Vector Search–optimized” profile on Azure Cloud
- Key facts and current state of the topic
- Elastic published guidance and a new Vector Search–optimized hardware profile for Elasticsearch on Azure, aimed at production ANN workloads (HNSW, ACORN‑style filtered search) and quantization (BBQ) with predictable memory/latency behavior. (elastic.co)
- Important context and background information
- Recent Lucene 10.x and Elasticsearch 9.1 gains (e.g., ACORN‑1 filtered HNSW and BBQ by default) materially improved filtered ANN and compression; a tuned infra profile helps teams realize those wins without bespoke capacity planning. (ir.elastic.co)
- Recent developments or changes
- The Azure‑specific profile packages instance sizing and storage guidance for vector‑heavy indices, reducing trial‑and‑error when rolling out compressed, filter‑heavy vector search or hybrid pipelines. Useful for late‑interaction or multi‑stage stacks that still rely on a Lucene/HNSW candidate stage. (elastic.co)
If you were expecting more: beyond this Elastic update, we did not find additional retrieval‑relevant releases or papers between October 16–23, 2025 that meet the bar for inclusion.