OpenSearch is an Apache 2.0-licensed fork of Elasticsearch 7.10.2 that AWS created in April 2021 after Elastic changed Elasticsearch's license from Apache 2.0 to dual SSPL / Elastic License v2. As of 2026 the two projects have diverged for nearly five years. Elasticsearch added AGPLv3 to its license options in August 2024 and continues to ship advanced features tied to paid subscriptions; OpenSearch was donated to the Linux Foundation as the OpenSearch Software Foundation in September 2024 and bundles features that Elasticsearch puts behind paid tiers. The right choice in 2026 turns on licensing requirements, feature needs, AWS commitment, and the team's preferred support model.
Quick Comparison
| Dimension | Elasticsearch (9.x, 2026) | OpenSearch (2.x, 2026) |
|---|---|---|
| License | AGPLv3 / ELv2 / SSPL (triple-licensed since Aug 2024) | Apache 2.0 |
| Governance | Elastic N.V. | OpenSearch Software Foundation (Linux Foundation) |
| Origin | Open-source 2010, license change 2021, AGPLv3 added 2024 | Forked from Elasticsearch 7.10.2 in 2021 |
| Latest major | 9.x | 2.x |
| Core engine | Apache Lucene 9.x/10.x | Apache Lucene 9.x |
| Query DSL | Elasticsearch DSL (proprietary extensions) | Compatible with ES 7.10 DSL, plus extensions |
| SQL support | Elastic SQL (subscription required for some features) | Built-in SQL plugin |
| Vector search | Native dense vector field, kNN, BBQ | k-NN plugin with FAISS / NMSLIB / Lucene engines |
| Machine learning | Built-in, subscription-gated | Built-in, free (ML Commons) |
| Security (RBAC, TLS, SAML, FLS/DLS) | Free in basic since 2021, advanced in paid tiers | Free, all features |
| Cross-cluster replication | Platinum+ | Free plugin |
| Alerting | Free since 8.x; advanced features paid | Free plugin |
| APM / Observability | Elastic Observability, paid tiers | OpenSearch Observability, free |
| Managed cloud option | Elastic Cloud | Amazon OpenSearch Service, Aiven, Bonsai |
| Client compatibility | Elasticsearch clients only (8.x+ strict) | OpenSearch clients; some ES 7.x clients work |
The 2021 Fork: What Actually Happened
In January 2021, Elastic announced it would change Elasticsearch and Kibana licenses from Apache 2.0 to a dual SSPL / Elastic License v2 model. The change took effect at version 7.11. SSPL is a copyleft license that the Open Source Initiative does not consider open source because it requires anyone offering the software as a service to release their full management stack under the same license - an explicit move to prevent cloud providers from monetizing Elasticsearch without contributing back.
AWS forked the last Apache 2.0 release (Elasticsearch 7.10.2) and renamed it OpenSearch. OpenSearch 1.0 shipped in July 2021. AWS's existing managed service, Amazon Elasticsearch Service, was rebranded to Amazon OpenSearch Service in September 2021.
In September 2024, AWS transferred governance of OpenSearch to the newly-formed OpenSearch Software Foundation under the Linux Foundation. Steering members include AWS, SAP, Uber, and others.
The 2024 License Change Back
In August 2024, Elastic added AGPLv3 as a third license option for Elasticsearch and Kibana, alongside ELv2 and SSPL. AGPLv3 is OSI-approved. Elastic's stated motivation was to restore the "open source" label without giving up the SSPL/ELv2 protections that prevent unmonitored cloud reselling. The codebases are now triple-licensed: users pick whichever of the three terms they want to use.
This change does not undo the OpenSearch fork - the projects have diverged for years and now have distinct features, governance, and roadmaps. It does mean newer Elasticsearch versions can legitimately be called open source.
Architecture Differences
Both projects descend from the same Lucene-backed distributed architecture - nodes, indices, shards, replicas, primary/replica writes, dedicated master nodes. The architecturally significant divergences:
- Index format: Elasticsearch and OpenSearch can read each other's snapshots back to the 7.10.2 fork point, but cannot read snapshots from newer versions of the other project. Cross-project snapshot restore generally requires reindex.
- Type system: Both removed mapping types in their 8.x / 2.x lines.
- Storage engine: Elasticsearch added Better Binary Quantization (BBQ) for vector search; OpenSearch supports FAISS and NMSLIB through k-NN plugin.
- Aggregations and DSL: Largely compatible for 7.10-vintage features; each has added new aggregations independently since.
- Cluster coordination: Both use the same gossip-based coordination subsystem (introduced in ES 7.0).
Feature Comparison
Vector search
| Elasticsearch | OpenSearch | |
|---|---|---|
| Dense vector field | Native (since 8.x) | k-NN plugin |
| HNSW algorithm | Yes (Lucene) | Yes (Lucene engine), plus FAISS and NMSLIB |
| Quantization | Scalar, binary, BBQ | PQ via FAISS |
| Approximate kNN | Yes | Yes |
| Exact kNN | Yes | Yes |
| Hybrid (lexical + vector) | Reciprocal rank fusion since 8.9 | Hybrid query plugin |
Machine learning
| Elasticsearch | OpenSearch | |
|---|---|---|
| Anomaly detection | Platinum+ | Free (ML Commons) |
| Forecasting | Platinum+ | Free |
| Trained model deployment | Platinum+ for some models | Free |
| AIOps | Enterprise | Limited |
Security
| Elasticsearch | OpenSearch | |
|---|---|---|
| TLS, RBAC, API keys | Free (since 8.0) | Free |
| SAML, OIDC, Kerberos | Free | Free |
| Field/document-level security | Platinum+ | Free |
| Audit logging | Platinum+ | Free |
This is where OpenSearch consistently wins on cost: features that Elastic gates behind Platinum+ are baseline in OpenSearch.
Pricing and Licensing
Elasticsearch
- Self-managed: Free download (AGPLv3 / ELv2 / SSPL). Most features in basic free; advanced features require subscription via Elastic Subscriptions.
- Elastic Cloud Hosted: Standard, Gold, Platinum, Enterprise tiers. See Elastic Cloud Pricing Guide. Realistic production cost ranges from $1.5k/mo (modest) to $50k+/mo (mid-sized).
- Elastic Cloud Serverless: VCU-based pricing, decoupled from infrastructure.
OpenSearch
- Self-managed: Free (Apache 2.0). All features included.
- Amazon OpenSearch Service: Managed by AWS, billed per instance-hour plus storage.
- Aiven, Bonsai, Logit.io: Third-party managed offerings.
OpenSearch is essentially always cheaper at the license layer. Total cost of ownership depends on managed-service pricing and engineering time for operations.
Performance
Benchmarks at the engine level show small, workload-dependent differences. Elasticsearch tends to lead in:
- Vector search latency at large scale (BBQ optimization)
- Complex aggregation throughput
- ML inference on CPU
OpenSearch tends to lead in:
- Indexing throughput on certain workloads
- Concurrent search throughput on small-document indices
- Recovery time after node loss (concurrent segment search shipped in 2.x)
For most production workloads the engine performance difference is below 20% and dwarfed by tuning, hardware, and shard layout decisions.
When to Choose Each
Choose Elasticsearch when:
- You're already on Elastic Stack and the migration cost is high
- You need the latest vector search optimizations (BBQ) or advanced ML
- You want Elastic's commercial support and Search/Observability/Security solutions
- AGPLv3, ELv2, or SSPL is acceptable for your distribution model
- You're on Elastic Cloud and the managed experience is valuable
Choose OpenSearch when:
- Apache 2.0 licensing is a hard requirement
- You're heavily invested in AWS and Amazon OpenSearch Service fits operationally
- You need RBAC, FLS/DLS, alerting, and CCR without paying for Platinum
- You want a vendor-neutral, foundation-governed project
- Your team can run with community support or third-party managed services
Migration Considerations
Elasticsearch 7.x to OpenSearch
- Direct snapshot restore works from ES 7.x snapshots
- Most client libraries continue to work after switching endpoints
- Some Elastic-specific features (Lens, Maps, Elastic ML) have no OpenSearch equivalent
- Index template syntax and ILM policies are compatible
OpenSearch to Elasticsearch
- Generally requires reindex - newer index formats aren't bidirectionally compatible
- Elastic 8.x clients are strict about server signature, so transparent fall-back is harder
- License obligations apply once switched
Either direction
- The longer the project sits on one side, the more drift accumulates. Plan migrations during planned upgrade windows, not as emergency response.
Operating Both with Pulse
Pulse is engine-agnostic - it monitors Elasticsearch and OpenSearch clusters with the same agentic SRE and root-cause workflows. Teams running mixed fleets (legacy ES alongside new OpenSearch) use Pulse for unified health, shard sizing, JVM heap analysis, and alerting across both engines, avoiding the cost of two separate monitoring stacks. For teams choosing between the two, Pulse can also model expected cost differences before migration.
Frequently Asked Questions
Q: What is the difference between OpenSearch and Elasticsearch?
A: OpenSearch is an Apache 2.0 fork of Elasticsearch 7.10.2 that AWS created in 2021 after Elastic changed Elasticsearch's license to SSPL / Elastic License v2. The two projects share a common ancestor but have diverged on features, licensing, and governance for nearly five years.
Q: Is Elasticsearch open source in 2026?
A: Yes - since August 2024, Elasticsearch is triple-licensed under AGPLv3, ELv2, and SSPL. AGPLv3 is OSI-approved, restoring the "open source" label. Earlier versions (7.11 through 8.15) were SSPL / ELv2 only and were not OSI-recognized as open source.
Q: Who owns OpenSearch?
A: OpenSearch is governed by the OpenSearch Software Foundation, a Linux Foundation project, since September 2024. AWS donated the project to the foundation; steering membership includes AWS, SAP, Uber, and others.
Q: Can I migrate from Elasticsearch to OpenSearch?
A: Yes, especially from Elasticsearch 7.x. OpenSearch can restore snapshots from ES 7.x, and most clients work after switching endpoints. Migrating from Elasticsearch 8.x or 9.x is harder - newer features and index formats aren't compatible, so reindex is usually required.
Q: Is OpenSearch faster than Elasticsearch?
A: Benchmarks show small workload-dependent differences. OpenSearch tends to lead on indexing throughput and concurrent search on certain workloads; Elasticsearch tends to lead on vector search latency and complex aggregations. For most production workloads the engine-level difference is below 20%.
Q: Does AWS OpenSearch Service support all OpenSearch features?
A: Most features are supported. Some advanced plugins and the very latest OpenSearch versions take time to reach the managed service. Always check the current Amazon OpenSearch Service version notes for parity with upstream OpenSearch.
Q: Should I use OpenSearch or Elasticsearch for vector search in 2026?
A: Elasticsearch has the edge in 2026 for high-scale vector workloads thanks to BBQ quantization and tighter Lucene integration. OpenSearch supports FAISS and NMSLIB via the k-NN plugin and is comparable for moderate scales. The decision usually comes down to licensing and ecosystem fit rather than vector performance alone.
Related Reading
- Elastic Subscriptions for Elasticsearch: Elasticsearch licensing tiers
- Elastic Cloud Pricing Guide: Elastic Cloud cost breakdown
- Managed Elasticsearch: Managed Elasticsearch and OpenSearch options
- OpenSearch Documentation: OpenSearch KB index
- Elasticsearch Documentation: Elasticsearch KB index
- Where to Download Elasticsearch: Get the distribution you need
