Elasticsearch vs OpenSearch
Elasticsearch vs OpenSearch after Elastic's AGPL licence option and OpenSearch's move to the Linux Foundation: features, vector search, hosting costs and migration.
Quick Answer
Choose Elasticsearch if you want the most feature-rich engine (ES|QL, mature ML, Elastic's observability and security apps) and are happy with AGPL, SSPL or Elastic's own licence. Choose OpenSearch if you want Apache 2.0 with foundation governance, already run on AWS, or need security features without paid tiers.
Reviewed by TechLogHub Engineering Team. Last updated October 15, 2026. Reflects Elastic's AGPLv3 licence option, Elasticsearch 9.x, OpenSearch 3.x under the Linux Foundation, and Elastic Cloud Serverless pricing updated in late 2025.
| Feature | Elasticsearch | OpenSearch |
|---|---|---|
| Licence | AGPLv3, SSPL or Elastic License 2.0 | Apache 2.0 |
| Governance | Single vendor (Elastic N.V.) | OpenSearch Software Foundation (Linux Foundation) |
| Current Major Version | 9.x (9.0 released April 2025) | 3.x (3.0 released May 2025) |
| Managed Services | Elastic Cloud on AWS, Azure, GCP | Amazon OpenSearch Service, Aiven, Instaclustr and others |
| Serverless Option | Elastic Cloud Serverless | Amazon OpenSearch Serverless |
| Vector Search | Yes, dense/sparse, hybrid, quantization | Yes, k-NN plugin (Faiss, Lucene engines) |
| Query Languages | Query DSL, ES|QL, SQL, EQL | Query DSL, PPL, SQL |
| Dashboards UI | Kibana | OpenSearch Dashboards |
| Security in Free Tier | Basic TLS and RBAC; advanced features paid | Full security plugin free |
| Client Compatibility | Elasticsearch clients only | OpenSearch clients (7.10-era ES clients partly) |
Elasticsearch
The original distributed search and analytics engine built on Apache Lucene, developed by Elastic and offered self-managed or on Elastic Cloud.
Pros
- Open source again: AGPLv3 was added as a licence option in August 2024 alongside SSPL and ELv2
- Richest feature set, including ES|QL, mature ML jobs and built-in observability and SIEM apps
- Strong vector and hybrid search with integrated inference and reranking
- Elastic Cloud Serverless removes cluster sizing, billed by compute units and retention
- Largest community, documentation base and third-party integrations
- Single vendor drives a coherent roadmap across Elasticsearch, Kibana and agents
Cons
- Advanced features such as SSO, field-level security and some ML need paid subscriptions
- AGPL copyleft can be a concern for teams embedding or redistributing it
- Elastic Cloud can cost more than equivalent Amazon OpenSearch Service capacity
- Licence history (2021 move away from Apache 2.0) left some users wary
- Recent clients and APIs no longer work against OpenSearch
Best For
Teams that want the most capable search, observability and security stack and value Elastic's integrated tooling.
OpenSearch
An Apache 2.0 fork of Elasticsearch and Kibana 7.10, created by AWS in 2021 and governed since September 2024 by the OpenSearch Software Foundation under the Linux Foundation.
Pros
- Permissive Apache 2.0 licence with no copyleft or service restrictions
- Vendor-neutral governance under the Linux Foundation
- Security plugin (fine-grained access control, audit logging) included free
- OpenSearch 3.x moved to Lucene 10 and added GPU-accelerated vector indexing (experimental)
- First-class managed option on AWS, including Amazon OpenSearch Serverless
- Wide choice of third-party hosts such as Aiven and Instaclustr
Cons
- Feature depth in ML, observability and SIEM trails Elastic's offerings
- AWS remains the dominant contributor despite foundation governance
- Smaller plugin and integration ecosystem than Elasticsearch
- APIs have diverged since 7.10, so tooling built for modern Elasticsearch may not work
- Documentation and community answers are thinner
Best For
Teams that want a permissively licensed search engine, run on AWS, or need security features without a subscription.
Licensing after Elastic's AGPL move
In 2021 Elastic moved Elasticsearch and Kibana from Apache 2.0 to a dual SSPL and Elastic License 2.0 model, mainly to stop cloud providers reselling it as a service. AWS responded by forking version 7.10 as OpenSearch. In August 2024 Elastic added the OSI-approved GNU AGPLv3 as a third option for the free source code, so Elasticsearch is once again open source by the OSI definition. AGPL is copyleft: if you modify Elasticsearch and make it available to users over a network, you must publish those modifications. For most teams running an unmodified cluster behind their own application this has no practical effect, but legal departments at some companies still prohibit AGPL components. Commercial features remain under Elastic's own licence and paid subscriptions. OpenSearch stays on Apache 2.0, a permissive licence with no copyleft and no restrictions on offering it as a service. In September 2024 the project moved to the OpenSearch Software Foundation under the Linux Foundation, which formalised multi-vendor governance, although AWS remains the largest contributor. If your organisation has a strict permissive-only policy, OpenSearch is the straightforward answer; otherwise licensing is no longer the decisive factor it was in 2021.
Feature gap: query languages, ML and vector search
The codebases have diverged for more than four years and each has added its own features. Elasticsearch offers ES|QL, a piped query language that has become central to Kibana, alongside the classic Query DSL, plus mature anomaly-detection ML, an integrated inference API and polished hybrid search with reranking. Elastic also ships full observability and SIEM products on top of the engine. OpenSearch has PPL and SQL plugins, an anomaly-detection plugin, a security analytics plugin and an ML Commons framework for connecting to embedding and LLM providers. OpenSearch 3.0, released in May 2025, moved to Lucene 10 and Java 21, added experimental GPU-accelerated vector index builds, native MCP support for AI agents, and experimental gRPC transport and pull-based ingestion from Kafka. Benchmarks between the two flip-flop by workload and version, and vendor-run benchmarks are rarely neutral. For vector search, both support approximate k-NN with HNSW and quantisation; OpenSearch also lets you choose the Faiss engine. In practice, Elasticsearch leads on breadth and polish, while OpenSearch is catching up fast on vector and AI-agent features. Prototype your own queries on both if the gap matters.
Managed hosting and cost
Elastic Cloud runs on AWS, Azure and Google Cloud in two forms: hosted deployments where you choose node sizes, and Elastic Cloud Serverless, which bills Elasticsearch projects by ingest, search and ML compute units plus retained data, and bills observability and security projects by gigabytes ingested and retained. Serverless removes shard and node tuning, which is the main operational pain of running either engine. OpenSearch's main managed home is Amazon OpenSearch Service, with provisioned domains and Amazon OpenSearch Serverless. Because it lives inside AWS, you can apply reserved instances and keep traffic inside your VPC. Other providers such as Aiven and Instaclustr also host OpenSearch on multiple clouds, which gives more competitive pressure on price than Elastic Cloud has. Self-hosting costs are similar for equivalent hardware, but OpenSearch includes fine-grained access control, audit logging and alerting at no cost, while Elasticsearch gates some of those behind paid tiers. For cost-sensitive teams on AWS, OpenSearch is usually cheaper end to end. For teams that would otherwise buy separate observability or SIEM tools, Elastic's bundled solutions can offset its higher price.
Migration and API compatibility
OpenSearch began as a drop-in replacement for Elasticsearch 7.10, and snapshots and most REST calls from that era still work. That compatibility has eroded with every release. Elasticsearch 8 and 9 changed APIs, defaults and security behaviour, and their official clients refuse to talk to OpenSearch; OpenSearch maintains its own clients for major languages. Moving from Elasticsearch 7.x to OpenSearch is usually straightforward: restore a snapshot or reindex, swap clients and test queries. Moving from Elasticsearch 8 or 9 is harder because newer index features, ES|QL queries and Kibana dashboards do not port directly; plan for reindexing and rebuilding visualisations. Going the other way, from OpenSearch to Elasticsearch, involves similar reindexing and replacing plugin-specific features such as PPL or the security plugin configuration. Abstraction helps. Keep search logic behind your own service layer, avoid engine-specific query features you do not need, and store source data so indexes can be rebuilt. If you are starting a new project, pick one engine deliberately and do not count on easy switching later, because the two projects are now separate products with distinct roadmaps.
Verdict
Choose Elasticsearch if you want the most complete search and analytics platform, plan to use Elastic's observability or security solutions, and are comfortable with AGPL or a commercial subscription. Choose OpenSearch if licence permissiveness and vendor-neutral governance matter, you are already on AWS, or you want fine-grained security without paying for it. For plain application search both are excellent, so let hosting cost and licence policy decide. Elastic's 2024 AGPL move removed the 'Elasticsearch isn't open source' argument, but it did not make the two projects interchangeable again.