We’re excited to announce the release of Weaviate v1.39.0!
This release brings a big batch of new capabilities and refinements: new ways to query Weaviate, a new quantization method, more control over your schema at runtime, and a broad round of performance work across search, replication and indexing.
Key Changes
Namespaces
Namespaces continue to mature, adding control-plane and data isolation between users on a shared cluster. This release adds namespace-local roles, namespace suspend support (RAFT states, endpoints and status checks for DB users), namespace graduation via backup/restore, and API key prefix visibility for admins on namespaced clusters.
RQ4
A new 4-bit Rotational Quantization option, together with improved SIMD vector search performance. RQ4 lets you push compression further while keeping search fast.
Hybrid MMR
Hybrid search now supports Maximal Marginal Relevance. You can add diversity to your hybrid results and avoid returning many near-identical objects at the top of the list.
Search REST API
A dedicated REST API for performing search queries, starting with POST /v1/search/{collection}/near-text. Responses come back in a consistent envelope containing id, properties, references and metadata.
gRPC Web
A new /grpc-web endpoint, making it possible to talk to Weaviate over gRPC from environments where plain gRPC isn’t available, such as the browser.
Alter Schema - Reindex Property (Preview)
Changing a property’s index type at runtime gets a reworked, resource-oriented API along with a set of reliability fixes covering recovery after restarts, class deletion racing an in-flight reindex, and task retry on schema-flip failures.
Alter Schema - Drop Vector Index (Preview)
Support for dropping indices from existing properties, so you can reclaim disk space by removing indices you no longer need. It comes with RBAC integration, multi-tenancy support (including cold tenants), and cleanup wired into segment compaction.
BM25 / BlockMax (BMW) Rework
Keyword search received a substantial overhaul: faster varint decoding, allocation cuts in the BlockMax WAND hot path, parallel block term creation across properties, and several correctness fixes around property boosts and duplicate properties. This release also adds support for cross-property AND matching in BM25 search.
Performance & Reliability
Async replication saw extensive optimization and hardening work: fewer allocations and less I/O during digest scans, batched hashtree pre-filtering for many-tenant clusters, and a number of fixes around read repair, tombstones and goroutine leaks. HFresh continues to improve with lower task dedup memory and locking fixes, replica movement now uses hard links to avoid halting compaction, and backups gained better error semantics, incremental backup fixes and configurable deduplication.
Modules & Integrations
New in this release: a generative-deepseek module and a multi2vec-twelvelabs multimodal vectorizer built on TwelveLabs Marengo. Existing modules also gained options, including location settings for text2vec-google and generative-google, endpoint configuration for the OpenAI client, and dimensions support for text2vec-aws.
Security & Observability Enhancements
Debug endpoints are now disabled by default, an SSRF bypass via X-BaseURL request headers was closed, and the base image and several dependencies were updated. On the observability side, this release adds unified background-process activity and duration metrics, plus persistent cluster and node identity for correlatable telemetry.
Breaking Changes
- None
For a comprehensive overview of all changes and improvements in this release, you can check out the full release notes here: Weaviate v1.39.0 Release Notes.
A big thank you to our new contributors in this release: @hashkanna, @vjsai, @VihaanAgarwal and @apoorva-01.
We encourage you to explore these new features and enhancements to see how they can benefit your projects:
Happy coding!