Elasticsearch vs OpenSearch for search and retrieval in 2026
OpenSearch is the Apache-2.0 fork of Elasticsearch 7.10.2, governed by the Linux Foundation's OpenSearch Software Foundation, while Elasticsearch is Elastic's engine offered under AGPLv3, SSPL or the Elastic License 2.0 with in-house embedding and reranker models, so the choice usually follows licence policy, the cloud a team runs on, and which vector features it needs.
Checked
OpenSearch began as a fork of the last Apache 2.0 release of Elasticsearch, 7.10.2, and reached production readiness with 1.0 in July 2021. AWS moved the project to the Linux Foundation's new OpenSearch Software Foundation in September 2024. Elasticsearch added AGPLv3 as a licence option in 2024, alongside SSPL and the Elastic License 2.0.
For a team building retrieval for agents, both engines offer mature BM25 keyword search and vector search in one place, which is what hybrid search needs. The differences are licence terms, where each is easiest to run, and the vector and model features each has added since the fork.
Licence and governance
Elasticsearch
The repository default is a choice of AGPLv3, SSPL v1 or Elastic License 2.0, and x-pack code is under the Elastic License 2.0 only. Elastic, the company, steers development.
OpenSearch
Apache License 2.0 throughout. Governed by the OpenSearch Software Foundation under the Linux Foundation, launched on 16 September 2024 with AWS, SAP and Uber as premier members.
Vector search and models
Elasticsearch
Dense and sparse vectors, BBQ quantization and the disk-based DiskBBQ format (shipped in 9.2 on 30 October 2025), the ELSER sparse model, and Jina embedding and reranker models after Elastic completed its acquisition of Jina AI on 9 October 2025.
OpenSearch
A k-NN vector engine on Faiss and Lucene, hybrid search and neural plugins. OpenSearch 3.0, generally available on 6 May 2025, added experimental GPU index builds. On AWS, vectors can be offloaded to Amazon S3 Vectors.
Which fits which team
Elasticsearch fits
Teams that already run Elastic, want embedding and reranking models from the same vendor as the engine, and are comfortable with one of its three licence options.
OpenSearch fits
Teams whose policy requires an Apache 2.0 licence, and teams on AWS that want Amazon OpenSearch Service and S3 Vectors in the same account.
Either way
Anthropic's Contextual Retrieval study found that adding BM25 to contextual embeddings, then a reranker, cut top-20 retrieval failures by 67% against its baseline. Both engines can run that keyword plus vector pattern.
How Motiontic approaches it
Motiontic's engineers are vendor-neutral. They load a sample of the client's own documents into the candidate stores, run the real questions through each with the same embedding model and reranker, compare recall, latency and cost per query against an eval set, and hand that eval set over so a later migration is a measured decision.
Elasticsearch and OpenSearch, figures retrieved 30 October 2026
| Elasticsearch | OpenSearch | |
|---|---|---|
| Origin | Elastic's original engine | Fork of Elasticsearch 7.10.2; 1.0 in July 2021 |
| Licence | AGPLv3, SSPL or Elastic License 2.0; x-pack ELv2 only | Apache 2.0 |
| Governance | Elastic (company) | OpenSearch Software Foundation, Linux Foundation |
| Latest release | 9.5.5 (6 Oct 2026) | 3.9.0 (29 Sep 2026) |
| Vector engine | Dense and sparse vectors, BBQ and DiskBBQ quantization | k-NN on Faiss and Lucene, experimental GPU index builds |
| First-party models | ELSER, Jina embeddings and rerankers | Neural plugins for external models |
| Managed options | Elastic Cloud, hosted and serverless | Amazon OpenSearch Service, managed and serverless |
| GitHub stars | About 78k | About 14k |
Common questions
Is OpenSearch a fork of Elasticsearch?
Yes. OpenSearch is derived from Elasticsearch 7.10.2, the last Apache 2.0 release, and became production-ready with 1.0 in July 2021.
Is Elasticsearch open source again?
Elastic added AGPLv3, an OSI-approved licence, as an option in 2024. The repository default is now a choice of AGPLv3, SSPL or the Elastic License 2.0, with x-pack code under the Elastic License 2.0 only.
Who governs OpenSearch?
The OpenSearch Software Foundation, part of the Linux Foundation, launched on 16 September 2024 when AWS moved the project there.
Which is better suited to vector search, Elasticsearch or OpenSearch?
Both support vector and hybrid search. Elasticsearch adds in-house embedding and reranker models and the DiskBBQ format. OpenSearch offers GPU index builds and S3 Vectors offload on AWS. A test on the team's own data and queries settles it.
Related
AI agent development
Agents that do real work for users, with the evals, traces and guardrails to keep them in production.
pgvector vs Pinecone
A Postgres extension compared with a managed serverless vector database.
Hybrid search
Keyword ranking and vector similarity, fused into one result list.
Reranker
A second pass that rereads the shortlist and puts the right passages first.
Vector quantization
Fewer bits per dimension: smaller, cheaper indexes at a small cost in recall.
Sources
- Elasticsearch LICENSE.txt
- OpenSearch FAQ (fork of Elasticsearch 7.10.2, Apache 2.0)
- Linux Foundation, OpenSearch Software Foundation (2024-09-16)
- Elastic, acquisition of Jina AI completed (2025-10-09)
- Elastic, DiskBBQ vector storage format (2025-10-30)
- OpenSearch 3.0 vector performance announcement
- AWS, Amazon S3 Vectors GA (2025-12-02)
- Anthropic, Contextual Retrieval (2024-09-19)
Running agents that have to hold up?
Describe the agent, the systems it touches and what it must never get wrong. The first call is with a lab engineer, who will say plainly whether it is ready for production, and whether Motiontic is the right team.