Qdrant

Qdrant 1.20 adds BM25 scoring directly on the full-text payload index, new turbo8/turbo16 vector datatypes, per-element nested payload indexes, and substring filter conditions, alongside faster HNSW builds, quantization search, and snapshot transfers. It also enforces internal cluster authentication by default when an API key is configured, so clusters that previously relied on an API key without enabling enforce_internal_auth must explicitly set this option or accept the new default before upgrading.

Matters for: Teams running multi-node Qdrant clusters with API key authentication and teams building search features that need BM25-ranked full-text retrieval.

Qdrant 1.19 lets you store vectors in 4-bit TurboQuant format as the primary storage to cut disk usage, set per-collection-component memory behavior (cold/cached/pinned), match filter keywords by prefix, and enforce resource limits through a new global quota API replacing the old max_resident_memory_percent setting. This release also fixes a path traversal vulnerability in S3-based snapshot handling and numerous data races and crashes in resharding, WAL replay, and payload indexing.

Matters for: Teams running large-scale Qdrant deployments with tight disk or memory budgets, and those using S3-based snapshot storage for backups.