Elasticsearch ships search innovation faster — ES|QL, the vector and semantic-search work, the integration catalogue — and the AGPL option ended the "not open source" era. OpenSearch gives you the full stack, security features included, under Apache 2.0 and neutral governance, with AWS’s managed service behind it. Deep on AWS or strict about permissive licensing: OpenSearch. Search quality as a product differentiator: Elasticsearch.
OpenSearch
The Apache-2.0 fork, now under the Linux Foundation.
Both descend from the same codebase: when Elastic relicensed Elasticsearch in early 2021, AWS forked version 7.10 and called it OpenSearch. Five years of independent development later, the Elasticsearch vs OpenSearch choice is no longer "same thing, different license" — the APIs have drifted, the feature sets have diverged, and both projects changed their governance story along the way.
Quick takes
If you're…
You run on AWS and want the path of least resistance→OpenSearchAmazon OpenSearch Service is the native managed offering, wired into CloudWatch, Firehose, and IAM.
Search relevance is your product (e-commerce, discovery)→ElasticsearchElastic’s relevance tooling — hybrid retrieval, semantic_text, reranking — moves faster.
Legal requires permissive open source across the stack→OpenSearchApache 2.0 under the OpenSearch Software Foundation, no copyleft, no vendor licenses.
You want SIEM/security analytics without per-feature pricing→OpenSearchSecurity analytics, alerting, and anomaly detection ship free in the distribution.
Your team lives in Kibana and the Elastic agent ecosystem→ElasticsearchOpenSearch Dashboards forked Kibana 7.10 and the muscle memory transfers only partially.
You need ES|QL-style piped queries for investigations→ElasticsearchES|QL is Elasticsearch-only; OpenSearch’s PPL is similar in spirit, different in syntax and depth.
You are choosing a vector store for RAG on AWS infrastructure→OpenSearchOpenSearch k-NN (Faiss/Lucene engines) is mature and is what Bedrock knowledge bases use underneath.
You want commercial support from the people who write the engine→ElasticsearchElastic sells exactly that; OpenSearch support comes from AWS or third parties.
Apache 2.0, full stop, for the engine, Dashboards, and the plugins. Governance moved from AWS to the OpenSearch Software Foundation under the Linux Foundation in September 2024, with SAP, Uber, Aiven, and Canonical alongside AWS. The neutrality is structural now, not promised.
Elasticsearch
Triple-licensed: AGPLv3 (added August 2024, OSI-approved), Elastic License 2.0, or SSPLv1. Open source again in the strict sense, but AGPL copyleft matters if you modify and serve it, and the trademark and roadmap remain one company’s.
core
Shared lineage, drifting APIs
tie
OpenSearch
Forked from Elasticsearch 7.10.2 and Kibana 7.10.2. The 7.x-era REST APIs, query DSL, and index concepts still feel identical. Everything added since — new APIs, settings, index features — is OpenSearch’s own and Elastic-incompatible.
Elasticsearch
Continued through 8.x and 9.x with its own breaking changes, new index formats, and features. There is no supported migration in either direction beyond the 7.10 ancestor: moving means reindexing, not upgrading.
features
Feature pace since the fork
edge: Elasticsearch
OpenSearch
Steady, with strengths in the analytics suite: security analytics, anomaly detection, index management, PPL query language, and serious k-NN work (Faiss and Lucene engines, GPU-accelerated index builds in the 3.x line). The cadence is real but the search-core innovation trails.
Elasticsearch
The faster shipper on the search core: ES|QL as a new query language, dense-vector improvements with int8/binary quantisation, semantic_text for managed embedding pipelines, retrievers for hybrid ranking, and the stateless serverless re-architecture. Elastic’s R&D depth on Lucene itself still shows.
features
What ships free
edge: OpenSearch
OpenSearch
Everything in the distribution: fine-grained security, SIEM-style security analytics, alerting, anomaly detection, cross-cluster replication, snapshots — all Apache 2.0. The project has no paid tier to protect, which shapes what lands in the box.
Elasticsearch
The free tier (Basic) is generous — security, vectors, much of the platform — but features like machine learning, some alerting connectors, and advanced tiers sit behind paid subscriptions. The line moves between releases; check it for the features you need.
ecosystem
Clients and library compatibility
depends
OpenSearch
Maintains forked clients (opensearch-py, opensearch-java, and so on) plus language-level compatibility with 7.10-era Elastic clients. Tooling that hardcodes Elastic version checks — some Beats versions, newer Logstash outputs — needs the OpenSearch outputs or Data Prepper instead.
Elasticsearch
Official Elastic clients from the 7.13/7.14 era onward perform a product check and refuse to connect to non-Elastic servers — the classic gotcha that breaks "drop-in" assumptions in both directions. Within its own ecosystem, client quality and codegen are excellent.
ecosystem
Managed offerings
depends
OpenSearch
Amazon OpenSearch Service (including serverless) is the flagship, deeply wired into the AWS fabric; Aiven, Instaclustr, DigitalOcean, and Oracle run it too. If your data already flows through AWS, the integration gravity is strong.
Elasticsearch
Elastic Cloud runs on AWS, GCP, and Azure, sells through the marketplaces, and gets new features first — including the serverless architecture. First-party only, but genuinely multi-cloud.
core
Performance claims
depends
OpenSearch
Has closed real gaps since the fork — concurrent segment search, k-NN improvements, faster aggregations — and publishes its own OpenSearch Benchmark suite. Independent third-party numbers are scarce.
Elasticsearch
Elastic publishes benchmarks claiming multi-x advantages over OpenSearch; OpenSearch contests the methodology. Both vendors benchmark their best case with their own tools (Rally vs OpenSearch Benchmark). Treat every number here as marketing until you run your own workload.
ecosystem
Ecosystem weight
edge: Elasticsearch
OpenSearch
Growing — 200+ maintainers, foundation members shipping in production (Uber, SAP), and the AWS integration surface. But the long tail of tutorials, plugins, certified connectors, and hireable experience still mostly says "Elasticsearch".
Elasticsearch
Fifteen years of accumulated ecosystem: agents and Beats, hundreds of integrations, books, courses, and the largest pool of operational lore in search. When something breaks at 3 a.m., the searchable answer base is unmatched.
History
The fork, and the licensing whiplash since.
Four dates explain most of the confusion in this comparison.
January 2021: Elastic relicensed Elasticsearch and Kibana from Apache 2.0 to SSPL/Elastic License with version 7.11, aimed squarely at AWS selling managed Elasticsearch. April 2021: AWS responded by forking the last Apache-2.0 release, 7.10.2, as OpenSearch. August 2024: Elastic added AGPLv3 as a third license option — Elasticsearch became OSI-approved open source again. September 2024: AWS transferred OpenSearch to the Linux Foundation, creating the OpenSearch Software Foundation and removing the "it’s just AWS’s fork" objection.
Both projects, in other words, fixed their original weakness. Elasticsearch is no longer proprietary-ish; OpenSearch is no longer single-vendor. What neither fixed is the split itself: indices, plugins, and clients stopped being interchangeable at 7.10, and five years of independent releases have made the gap permanent. Anyone promising you a trivial migration between them in either direction is describing 2021.
Practical consequence: treat them as two products that share ancestry and a query DSL dialect, the way MariaDB and MySQL share one. Choose on governance, cloud alignment, and which feature set maps to your workload — not on the assumption that you can switch later with a config change.
OpenSearch Service plus free security analytics, fed by Firehose and CloudWatch. The integration gravity and the pricing both point the same way.
E-commerce or content search where relevance moves revenue
Elasticsearch
Hybrid retrieval, semantic_text, reranking, ES|QL — the relevance toolchain is deeper and improving faster.
Open-source-only policy, self-hosted
OpenSearch
Apache 2.0 across engine, dashboards, and security features, governed by a foundation. No license lawyer required.
Established Elastic stack, wondering whether to jump
Elasticsearch
The AGPL option removed the main reason teams fled in 2021. A migration means reindexing, retooling dashboards, and retraining — pay that only for a concrete win.
Vector search backing a RAG feature on AWS
OpenSearch
OpenSearch k-NN is mature, Bedrock-integrated, and serverless if you want it. Elastic’s vector work is excellent too — this pick is about platform fit.
Mostly log search, cost is the complaint
Quickwit or ClickHouse
If you never use relevance scoring, a search cluster is an expensive way to grep. Object-storage-native log search or a columnar store cuts the bill structurally.