Open source, self-hosted - verified
OpenSearch (k-NN, open source)
OpenSearch is listed under open source, self-hosted (deployment type: open source self-host). Open source; no price. Max dimensions: 16,000. Read from the vendor's own pages on October 18, 2026.
Read from docs.opensearch.org, raw.githubusercontent.com. Docs - License - Compare every vector database
Status
- Availability label
- Not stated
No release-status label is quoted from the pages checked.
How OpenSearch charges
Every price OpenSearch states, in its own unit, with the quote and where it comes from. Nothing is converted.
Open source; no price.
Free tier and limits
- Free tier
- Not stated
Open-source software; no hosted free tier applies.
- Limits
- Not stated
No vector-count limit is stated on the pages checked.
Search and indexing
- Max dimensions
16,000
“dimension Integer The size of the vectors used. Valid values are in the [1, 16,000] range.”
docs.opensearch.org - Distance metrics
l1, l2, linf, cosinesimil, innerproduct, hamming, hammingbit
Note: Not every method/engine combination supports every space
“Valid values are l1, l2, linf, cosinesimil, innerproduct, hamming, and hammingbit.”
docs.opensearch.org - Index types
HNSW (Lucene, Faiss), IVF (Faiss), flat (Lucene, OpenSearch 3.6+)
“OpenSearch supports the following engines: - Lucene: The native search library, offering an HNSW implementation with efficient filtering capabilities - Faiss (Facebook AI Similarity Search): A comprehensive library implementing both the HNSW and IVF methods”
docs.opensearch.org “flat No l2, cosinesimil, innerproduct (supported in OpenSearch 3.6 and later)”
docs.opensearch.org - Hybrid and sparse search
- Not stated
Not described on the k-NN field type or methods pages checked.
- Metadata filtering
Yes (efficient filtering with the Lucene engine)
“Lucene: The native search library, offering an HNSW implementation with efficient filtering capabilities”
docs.opensearch.org - Multi-tenancy
- Not stated
Not described on the pages checked.
Built-in embedding and reranking
- Built-in embedding
- Not stated
Not stated on the pages checked.
- Built-in reranking
- Not stated
Not stated on the pages checked.
Deployment, license and SDKs
- Regions and clouds
- Not stated
Self-hosted software; regions do not apply.
- BYOC and self-host
Open source (Apache-2.0)
“Apache License Version 2.0, January 2004”
raw.githubusercontent.com - License
Apache-2.0 (OpenSearch and the k-NN plugin)
“Apache License Version 2.0, January 2004”
raw.githubusercontent.com “Apache License Version 2.0, January 2004”
raw.githubusercontent.com - SDKs
- Not stated
Not listed on the pages checked.
Vendor-stated claims
Performance, scale and SLA figures as OpenSearch states them. Not measured.
No performance, scale or SLA figure is recorded from the pages checked.
Research notes
The docs list a flat Lucene method as supported from OpenSearch 3.6.
FAQ
How much does OpenSearch cost?
Open source; no price.
What is the maximum vector dimension in OpenSearch?
16,000, per the vendor: "dimension Integer The size of the vectors used. Valid values are in the [1, 16,000] range."
Does OpenSearch support hybrid search?
Not stated. Not described on the k-NN field type or methods pages checked.
Does OpenSearch have built-in embedding or reranking?
Embedding: not stated. Not stated on the pages checked. Reranking: not stated. Not stated on the pages checked.
Sources
- Source: docs.opensearch.org/latest/mappings/supported-field-types/knn-methods-engines
- Source: docs.opensearch.org/latest/mappings/supported-field-types/knn-vector
- Source: raw.githubusercontent.com/opensearch-project/OpenSearch/main/LICENSE.txt
- Source: raw.githubusercontent.com/opensearch-project/k-NN/main/LICENSE.txt
Last verified from each vendor's own docs, pricing pages, READMEs and license files; every quote was string-matched against the page that day. This is a dated snapshot, not a benchmark: prices are shown in each product's own unit and change without notice, so check the linked source before you commit.
Newsletter
Know when a vector database changes its pricing
One email per week. New model releases, retrieval tooling and pricing changes and what they mean for what you build.