AI is reshaping how we build and run modern applications. From real-time recommendations to agentic assistants, teams need data platforms that can keep up with new performance and flexibility demands. That’s what Couchbase 8.0 is built for—a unified platform that brings operational, analytical, and vector-based workloads together so developers can build faster, smarter, and more cost-effective AI-powered applications.
General availability of Couchbase Server 8.0
Today we introduce Couchbase Server 8.0, our latest release for self-managed and fully-managed Capella deployments. With over 400 features and changes, Couchbase 8.0 delivers breakthrough innovations in vector indexing, vector search usage and performance, and cluster security, scalability and reliability. These new features help transform Couchbase into the AI data backbone needed for tomorrow’s generation of AI-powered applications and agentic systems.
Agentic systems are operational applications
We have long argued that agentic systems are best viewed as operational applications because they require the responsiveness, availability, distributed scale, and performance of platforms like Couchbase and Capella. And we have historically contended that assembling an operational application that is powered by a collection of purpose-built databases is a bad idea. Doing that with AI could be a disaster.
Check out the best multipurpose vector database
Today, we’re adding another category of database functionality that should be built-into a multipurpose platform, not live alongside one. Couchbase 8.0 becomes the best, most versatile vector database, along with being a fantastic JSON, KV-caching, search, eventing, mobile, and analytic multipurpose database platform.
Vector search that scales to a billion—and beyond
AI-driven applications depend on finding the right context instantly. That means fast, accurate, vector retrieval at massive scale. With the new Hyperscale Vector Index (HVI) in Couchbase 8.0, that’s now possible—without tradeoffs between speed, accuracy, or cost.
In independent billion-scale testing, HVI delivered up to 19,000 queries per second con 28 millisecond latency when adjusted for reasonable recall accuracy of 66%. Compared to a leading cloud database, Couchbase ran over 3,000 times faster. And when we turned up the adjustment for high recall accuracy (93% on modest hardware), Couchbase handled 350 times more queries per second.
Our new Hyperscale Vector Index, has already been tested to easily scale beyond one billion vectors with exceptional query throughput, recall accuracy, and millisecond latency. This not only helps customers improve accuracy and trust within their AI applications, it also helps make the use of GenAI more affordable. It will drive down the total cost of ownership of RAG and agentic use cases, especially when it is difficult to anticipate what users may ask of Large Language Models (LLMs).
Instead of using HNSW, IVF, or DiskANN, Hyperscale Vector Index is powered by a novel hybrid algorithm that combines the strengths of graph and cluster-based algorithms, based on Microsoft’s Vamana paper combined with IVF. The advantage to this design is that it utilizes both distributed in-memory and partitioned-on-disk processing, resulting in the best-in-class performance in terms of capacity, throughputs, latency, and recall. It is the preferred index to use when a large corpus of data must be vectorized while developers do not fully control the content supplied in prompts, such as chatbots. This implementation has many advantages, which we will explore in the future. But today, we just want to show it off.
Hyperscale vector index benchmark competition
In a fresh head-to-head vector performance benchmark between Couchbase and MongoDB Atlas, Couchbase’s new Hyperscale Vector Index achieved exceptional vector retrieval performance, measured in queries per second (QpS), against a common medium-sized vector set, and then also a billion vector dataset with 128 dimensions. The tests used the VDBBench methodology and toolkit, and measured queries per second (QpS), response latency in milliseconds, and recall accuracy percentage.
By varying the breadth of centroid clusters scanned (from 10 to 100), the tests are able to adjust retrieval performance and latency against their vector recall accuracy. Centroids are clusters of similar vectors. When scanning fewer centroids, queries per second (QpS) increase, but vector accuracy may be lower. Scanning more centroids improves accuracy, but may also increase latency.
The benchmark results demonstrate that Couchbase’s Hyperscale Vector Index can deliver just over 19,000 queries per second with a latency of only 28 milliseconds, when adjusted for lower accuracy (66%). This is 3,100 times faster than the same test and settings for MongoDB Atlas, which could only execute 6 queries per second at 57% recall accuracy.
When configured to favor recall accuracy, MongoDB’s performance dropped to 2 queries per second, and its latency responsiveness jumped to over 40 seconds. Couchbase performed over 700 QpS, with sub-second latency of 369 milliseconds. Atlas’ recall accuracy was 89% to Couchbase’s 93%. When operating at billion-vector scale, Couchbase’s Hyperscale Vector Index works harder, faster, smarter, and costs less.
The Hyperscale Vector Index is an extension of the original Index Service in Couchbase, and it inherits its existing deployment options, scale, distributed partitioning, and performance characteristics.
Composite Vector Index (CVI)
While we were at it, we added the Composite Vector Index for situations when developers want to define a prefiltered, narrowed, vector result set, also at millisecond speed.
Composite vector index is part of existing secondary index functions (GSI) through which you can build an index combining vector and other supported data types like strings, numbers and booleans. It helps narrow the focus of a vector request and is useful when developers control the contents of prompts within an LLM engagement. Thus it can apply filtering criteria before requesting specific vectors from Couchbase and minimize LLM token consumption without compromising accuracy.
Couchbase deploys vector search on premises, in Capella, and on mobile. Who else does that?
These new, massively scalable, vector indexing options are added to our existing, hybrid-vector search capabilities powered by our Search Service. Couchbase is now the only database platform to offer tres flexible, and highly scalable vector search options for self-managed systems on premises, in Kubernetes, and fully managed Capella deployments. Add to that our mobile vector search, and you can see how we can become the backbone for your AI applications that serve end users wherever they’re located in our AI world.
What else is in Couchbase 8.0?
Every service in Couchbase enjoys major enhancements. Let’s look at the changes for each Couchbase service:
Servicio de datos
Native encryption at rest with KMS integration for customer-managed keys. Data service is the first of each Couchbase service to be encrypted. Others like Query, Index and Search will follow in a subsequent release.
Includes centralized policy control with auto key rotation
90% Lower memory quota for Magma (100MB)
Smaller cluster map option of 128 vbuckets instead of 1024
Faster node activation as cache warms with new bucket warmup options (Background, Blocking, None)
Memcached bucket type is removed, deprecated since version 6.5
Servicio de consulta
Natural Language input for queries from Couchbase Server through command line shell, SQL++ and Query Workbench using the Capella iQ access credentials. Ask questions, with “USING AI” command or REST API commands starting with “natural_“.
Query workload repository and reporting maintains snapshots and reports to ease troubleshooting queries. A user-defined collection collects elapsed time, CPU and memory usage, KV fetch, executions and more.
Auto update optimization statistics for ideal query plan generation as query characteristics evolve
New SQL++ keywords and clauses for vector index creation including, CREATE VECTOR INDEX with optional INCLUDE, PARTICIÓN PORy DONDE clauses, plus extensions to the CON clause for vector-specific parameters such as, Dimension, Descripción, Similarity metric, Train_listy Num_replicas.
New SQL++ functions for vector selection, DISTANCIA_VECTOR_APROX
Vector Indexes are available through Query Workbench GUI, Capella UI, REST API to the Query Service, SDKs and via model frameworks like LlamaIndex and LangChain
Servicio de índices
New feature settings for vector index creation
Algorithms: IVF for GSI Composite, and IVF + Vamana (Hybrid) for hyperscale
SQL++: CREATE/ALTER/DROP INDEX through SQL++, REST API, and SDK
Quantization: Tuning index with choice of PQ, SQ variants for reduced memory usage
Similarity distance: Cosine, Dot Product, L2 and Euclidean for diverse application needs
Partitioned indexes: For scalability into multi-billion vectors and granular indexing requirements
New feature options for vector search
Simple search query: Basic ANN scans with vector fields in ORDENAR POR
Pre-filtering in Composite Index and Inline filtering in Hyperscale index with INCLUDE columns for reducing search space
Pushdowns to Indexer: For filtering and limiting documents to improve performances
Projections: Support for projections like vector distance
Reranking results: For improving recall with a performance tradeoff
Servicio de búsqueda
User-defined synonyms available to reference in search queries
Filter which documents to be indexed by search service
Best match (BM25) scoring for better hybrid search results
Read-replica partitions added to Search Service for faster query throughput
Search Vector Index performance has doubled through better SIMD (Single Instruction, Multiple Data) support using avx2 instruction set
Servicio de concursos
Eventing Service has been re-architected for scale, speed, and security with dramatic results
Set eventing options at the scope or bucket level of execution
Configure eventing service nodes by scope
TLS node to node encryption for internal communication
Administrador de clústeres
Auto-failover of ephemeral buckets and non-responsive disks
Adjust non-KV Multidimensional Scaling (MDS) services without introducing new target nodes
Aggregate SDK client metrics on cluster for easier monitoring and troubleshooting
Lock/unlock user accounts and monitor activity
Upgrade path requires version 7.2 or higher, earlier versions must upgrade to 7.2.3 first
Replicación entre centros de datos (XDCR)
New bucket property, “EnableCrossClusterVersioning” designed to enable:
Bi-directional replication with mobile buckets in Sync Gateway or Capella App Services
Target cluster awareness of inbound replications for easier management
Conflict logging for documents modified on both ends during conflict timeframe window
XDCR Diagnostic Utility to check data consistency between clusters
Backups
Point-in-time recovery preview before 8.1 GA
Reduce data loss window to user-defined timing from hours, to a few minutes, or even sub-seconds
Backup Retention Period and expiration settings to set expiration dates for backups
Auto-resolve naming conflicts with cbbackupmgr
Built for developers, trusted by enterprises
Couchbase 8.0 combines speed, scale, and flexibility in a single platform that runs anywhere—on-prem, in Capella DBaaS, or at the edge. It’s designed for the developers shaping tomorrow’s AI-powered experiences and for the enterprises that rely on them to run critical applications.
“Our customers can find relevant content based on meaning and context, not just exact keywords. As a Capella customer, we’re excited for Couchbase 8.0 and the scalability and TCO benefits that make it the ideal solution for our AI-powered video platform,” said Ian Merrington, CTO at Seenit.
Couchbase 8.0 is now generally available. Explore what’s new and see how teams are using it to build next-generation AI and agentic systems today.
Recibe actualizaciones del blog de Couchbase en tu bandeja de entrada
Autor
Publicado por Jeff Morris, Vicepresidente de Marketing de Producto
Jeff Morris es vicepresidente de marketing de productos y soluciones de Couchbase. Lleva más de tres décadas comercializando herramientas de desarrollo de software, bases de datos, herramientas analíticas, servicios en la nube y otros productos de código abierto. Él sería el primero en decir que cualquiera que busque una base de datos como servicio en la nube rápida, flexible, familiar y asequible puede dejar de buscar después de echar un vistazo a Couchbase.
