v2.6.6

Release date: November 21, 2025

| Milvus Version | Python SDK Version | Node.js SDK Version | Java SDK Version | Go SDK Version | |:-------------- |:------------------|:--------------------|:-----------------|:---------------| | 2.6.6 | 2.6.3 | 2.6.4 | 2.6.8 | 2.6.1 |

We are excited to announce the release of Milvus 2.6.6, featuring a range of powerful new capabilities, performance enhancements, and essential bug fixes. This update introduces important features such as Geospatial and Timestampz data type, Boost ranker for rescoring, etc. This release also has many crucial scalar filtering performance improvements. Several critical bugs have also been addressed to ensure greater stability and reliability. With this release, Milvus continues to provide a more robust and efficient experience for all users. Below are the key highlights of this release.

• Geospatial Data Type: Milvus introduces support for the Geometry data type, representing OGC-compliant geometric objects such as POINT, LINESTRING, and POLYGON. This type supports multiple spatial relationship operators (st_contains, st_intersects, st_within, st_dwithin, …) and provides an RTREE spatial index to accelerate spatial filtering and query execution. This enables efficient storage and querying of geospatial shapes for LBS, mapping, and other spatial workloads.
• Timestamptz Data Type: Milvus introduces the TIMESTAMPTZ data type, providing timezone awareness for all temporal data. This feature enables consistent data management across global deployments by allowing users to define a default time context using the timezone property on Databases and Collections. Crucially, the field fully supports expression-based filtering for time range queries, and retrieval operations (query and search) support a timezone parameter for instant, on-the-fly conversion of timestamps into the required local format upon output.
• Boost Ranker: Instead of relying solely on semantic similarity calculated based on vector distances, Boost Ranker allows Milvus to use the optional filtering condition within the function to find matches among search result candidates and boosts the scores of those matches by applying the specified weight, helping promote or demote the rankings of the matched entities in the final result.
• STL_SORT index now supports VARCHAR and TIMESTAMPTZ datatype.
• You may now enable dynamic field of an existing collection by altering it.
• Fixed cve-2025-63811.

Features

• Added new config and enabled dynamic update configs (#45363)

Improvements

• Fixed cve-2025-63811 (#45658)
• Removed large segment id arrays from querynode logs (#45720)
• Updated multiple places where the expr copied the input values in every loop (#45712)
• Optimized term expr performance (#45671)
• Prefetched vector chunks for sealed non-indexed segments (#45666)
• Expr: only prefetched chunks once (#45555)
• Added nullable support for geometry and timestamptz types (#45522)
• Increased session ttl from 10s to 30s (#45517)
• Added more metrics for ddl framework (#45559)

Bug fixes

• Fixed partial update panic with timestamptz (#45741)
• Used 2.6.6 for milvus ddl upgrading (#45739)
• Used latest timetick to expire cache (#45699)
• Made streamingnode exit when it failed initializing (#45732)
• Protected tbb concurrent_map emplace to avoid race condition deadlock (#45682)
• Prevented panic when streaming coord shutdown but query coord still worked (#45696)
• Set task init when worker didn’t have task (#45676)
• Prevented deadlock in runcomponent when prepare failed (#45647)
• Prevented panic when double closing channel of ack broadcast ()