# Milvus 3.0 Adds Lake-Native Vector Retrieval > Source: > Published: 2026-08-03 07:33:02+00:00 # Milvus 3.0 Adds Lake-Native Vector Retrieval The Milvus project released version 3.0.0 on July 29, 2026, extending its open-source vector database with lake-native retrieval over data held in open formats on object storage. The release adds external collections, online schema changes, sparse-index upgrades, faceted search, composable reranking, long-text fields, and broader storage-format support. The Milvus project released **Milvus 3.0.0** on July 29, 2026, according to its official release notes. The update extends the open-source vector database's lake-native architecture and moves more retrieval, ranking, and schema operations into the database. Independent coverage from TechTarget and Blocks & Files described the broader launch earlier in July; the official versioned documentation is the authority for the 3.0.0 release date and final feature set. ### Retrieval over lake-resident data External Collections let Milvus reference data held in formats such as Parquet, Iceberg, Lance, and Vortex, then build indexes and search without first copying the source table into a separate serving store. Version 3.0.0 expands that path so external fields can feed functions that generate BM25 sparse vectors, MinHash signatures, and text embeddings. It also adds incremental handling for newly added columns and a format that can expose Milvus snapshots as external tables. For teams already operating a data lake, this design can remove a synchronization surface and reduce duplicate storage. It does not remove the need to test object-store access patterns, indexing time, query latency, permissions, and failure recovery against a real workload. TechTarget noted that the approach is most useful when source data already lives in formats Milvus can read. ### More query work inside Milvus Milvus 3.0.0 adds online column add, backfill, and drop operations; a redesigned sparse index with SINDI and block-max algorithms; faceted search; composable Function Chain reranking; long-text fields; and FAISS index-factory passthrough. The release also expands nested StructArray search and supports Vortex and Lance as storage formats. The official notes say Storage V3 remains disabled by default, and features that depend on it require enabling common.storage.useLoonFFI. New index versions are also opt-in, and using features that change serialized data formats can prevent rollback to Milvus 2.6. Those compatibility details matter more to operators than the headline architecture: adoption requires a staged test of indexes, storage, access controls, rollback, and retrieval quality. For retrieval-augmented generation and agent systems, the practical value is the ability to combine dense and sparse retrieval, filtering, aggregation, faceting, and reranking closer to the data. The benefit should still be measured against corpus-specific relevance metrics and end-to-end cost rather than assumed from feature availability alone. ## Key Points - 1Milvus 3.0.0 can index and search lake-resident data in open formats without requiring a separate vector-serving copy. - 2The release adds online schema changes, sparse-index upgrades, faceting, reranking chains, long-text fields, and broader nested retrieval support. - 3Storage V3 and new index versions remain opt-in, so teams should test compatibility, rollback, latency, and retrieval quality before production adoption. ## Scoring Rationale Milvus 3.0.0 is a technically meaningful release for retrieval and RAG teams using open lake formats. Its production value depends on workload-specific validation because major storage and index paths remain opt-in and introduce compatibility considerations. ## Sources Primary source and supporting public references used for this report. Practice interview problems based on real data 1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)