OpenSearch vs Elasticsearch in 2026
OpenSearch began as a fork of Elasticsearch. Here is how the two differ on licence, vector search and migration, and how to choose between them.
45 minutes with an engineer: we review your workload and sketch the design.
- ✓Operated by AceMQ
- ✓Sourced, 30 Oct 2026
Elasticsearch vs OpenSearch: OpenSearch is the Apache 2.0 fork of Elasticsearch, and Elasticsearch is Elastic's own product under Elastic licences that gained an AGPL option in 2024. Choose OpenSearch for an Apache-licensed engine you can host anywhere. Choose Elasticsearch when you need Elastic's own features, support or cloud service.
- PostgreSQL
- MySQL
- MongoDB
- Redis
- Valkey
- OpenSearch
- Kafka
- RabbitMQ
- AWS
- Google Cloud
- Azure
- Kubernetes
- OpenShift
- Prometheus
- Grafana
- Datadog
- Dynatrace
- Splunk
OpenSearch and Elasticsearch side by side
OpenSearch and Elasticsearch share a history up to the fork in 2021 and differ in licence, governance and which features they have added since. The table gives the facts that matter for a hosting or procurement decision.
| Factor | OpenSearch | Elasticsearch |
|---|---|---|
| Licence | Apache License 2.0 (OpenSearch) | Elastic License 2.0 (ELv2) and SSPL; Elastic announced an AGPL option in August 2024 (Elastic) |
| Origin | Began as a fork of Elasticsearch and Kibana, announced in January 2021; version 1.0 shipped in July 2021 (OpenSearch) | Developed by Elastic, the original project |
| Vector search | k-NN with Lucene and Faiss engines; HNSW, and IVF in Faiss (OpenSearch docs) | dense_vector fields with approximate and exact kNN search (Elastic docs) |
| Keyword plus vector | A hybrid query that combines keyword and vector scores (OpenSearch docs) | Documented in Elastic's search docs; read the page for the version you plan to run |
| Moving between them | Migration Assistant moves data, metadata and live traffic from Elasticsearch (OpenSearch docs) | Not applicable: the documented route is into OpenSearch |
| Run by OSStend | Yes, see managed OpenSearch | No. OSStend runs OpenSearch; Elastic sells its own cloud service |
Sources: OpenSearch about, Elastic licence announcement, OpenSearch k-NN docs, Elastic kNN docs and Migration Assistant docs. Checked 30 Oct 2026. Vendor documentation changes; confirm on the linked pages.
How this comparison was made
This page compares what each project's own documentation and announcements state, read on 30 Oct 2026. It contains no benchmark run by OSStend and quotes no vendor benchmark.
Search benchmarks depend on hardware, mappings, query mix and data, so they rarely transfer to another workload. If speed decides the choice, test both engines on a copy of your own index. OSStend runs OpenSearch, so it has an interest in that side of the comparison; the sections below say where Elasticsearch is the better choice. More side-by-side pages are on the compare hub.
Licence and governance
OpenSearch is licensed under Apache 2.0, which has no restriction on hosting it as a service (OpenSearch). Elasticsearch has been offered under ELv2 and SSPL, and Elastic added AGPL as a third option from August 2024 (Elastic).
AGPL is an OSI-approved open source licence, so Elasticsearch is open source again for teams whose policy allows it. Some enterprises still prohibit AGPL, ELv2 or SSPL software outright, and a licence review decides that, not a feature table. Apache 2.0 is the permissive choice.
Features and compatibility since the fork
OpenSearch and Elasticsearch share core concepts such as indices, shards and the query DSL, because one came from the other. Each project has added features since the fork, so check every feature your applications use against both sets of documentation.
The risk sits in newer APIs, client libraries pinned to one vendor, and plugins or add-ons that only exist on one side. A review of the queries, mappings, ingest pipelines and plugins in use settles it before any move. Applications written against Elasticsearch versions near the fork are the easiest to test on OpenSearch.
Vector and hybrid search
Both engines can store embeddings and run nearest-neighbour search next to keyword search. OpenSearch documents k-NN with Lucene and Faiss engines and a hybrid query for combining keyword and vector scores (OpenSearch docs, checked 30 Oct 2026); Elasticsearch documents dense_vector fields with approximate and exact kNN (Elastic docs, checked 30 Oct 2026).
For retrieval in AI applications the engines are close enough that the deciding factors are usually licence, the cloud service you already use and the features you have built on. See vector database for RAG for how OpenSearch compares with pgvector for that job.
Moving from Elasticsearch to OpenSearch
OpenSearch documents a Migration Assistant that moves data, metadata and live traffic from Elasticsearch to OpenSearch (OpenSearch docs, checked 30 Oct 2026). It describes a snapshot-based backfill for a migration with planned downtime and a capture-and-replay route for zero downtime.
Inventory
List indices, mappings, queries, ingest pipelines, plugins and client libraries in use.
Build
Stand up OpenSearch to the signed-off design beside the existing cluster.
Copy and test
Backfill the data and run the application's tests against OpenSearch.
Cut over
Move clients in stages and keep the old cluster available for rollback.
The OSStend path is described on Elasticsearch to OpenSearch migration.
Where OSStend fits, and when it does not
OSStend fits when OpenSearch is the right engine and you want it sized for your index and run 24/7 in your cloud account, your data center or OSStend's cloud. It does not fit when Elastic's own product is what you need.
Contact us if
- ✓You want OpenSearch run for you, with dedicated nodes sized for your index.
- ✓You are leaving Elastic Cloud or Amazon OpenSearch Service and want the migration run for you. See the Elastic Cloud comparison.
- ✓The cluster must run in your own account or data center.
Not a fit if
- ✕You rely on a feature, add-on or support contract that only Elastic provides.
- ✕You want Elastic's own managed cloud service.
- ✕Your search workload is small and a turnkey plan covers it at lower cost.
Sizing, topology and pricing for OpenSearch are on the managed OpenSearch page.
OpenSearch and Elasticsearch questions
What is the difference between OpenSearch and Elasticsearch?
OpenSearch is an Apache 2.0 fork of Elasticsearch and Kibana, announced in January 2021. Elasticsearch is developed by Elastic under ELv2, SSPL and, since 2024, an AGPL option. They share core concepts and have added different features since the fork.
Is Elasticsearch open source again?
Elastic announced in August 2024 that AGPL, an OSI-approved open source licence, would be added as an option next to ELv2 and SSPL. Whether AGPL suits a given organisation depends on its own licence policy.
Can OpenSearch replace Elasticsearch?
For many workloads, yes. OpenSearch documents a Migration Assistant that moves data, metadata and live traffic from Elasticsearch. Queries, plugins and client libraries need testing first, because features have diverged since the fork.
Do both support vector search?
Yes. OpenSearch documents k-NN with Lucene and Faiss engines and a hybrid keyword-plus-vector query. Elasticsearch documents dense_vector fields with approximate and exact kNN search.
Related pages
- OpenSearchElasticsearch as a service on managed OpenSearchElasticsearch as a service on OpenSearch: dedicated master and data nodes, hybrid keyword and vector search and snapshots, sized for your index and run 24/7.
- ComparisonsComparisons: open-source engines and servicesComparisons of open-source engines and managed services: pgvector vs Pinecone, pgvector vs Qdrant, managed PostgreSQL pricing, Valkey vs Redis, OpenSearch vs Elasticsearch, PostgreSQL vs MySQL and more.
- AI data · RetrievalVector database for RAG on your own infrastructureA vector database for RAG on your own infrastructure: when PostgreSQL with pgvector is enough, when OpenSearch hybrid search fits, and how to size retrieval.
- Migration guideElasticsearch to OpenSearch migration: 3 methodsElasticsearch to OpenSearch migration: remote reindex, snapshot restore or Migration Assistant, the version limits that decide which, and how cutover works.
- AlternativeElastic Cloud pricing and alternatives in 2026Elastic Cloud pricing, the licence history and the alternative: managed OpenSearch under Apache 2.0, tailored to your workload and run where you choose.
Choose the search engine with an engineer
45 minutes: your queries and plugins are checked and the OpenSearch design is sketched.
- ✓A design sketch for your workload, not a plan menu
- ✓A zero-downtime cutover path for what you run today
- ✓A price for that design, after the review