Serverless OpenSearch
Serverless OpenSearch. No OCUs.
WarpSearch serves the OpenSearch API as a serverless service. Create a collection, point opensearch-py or any OpenSearch client at api.warpsearch.io, and search. No OCUs to size or keep awake, no clusters to run. You pay per GB and per request, from $10 a month, with a 99.9% uptime SLA on every paid account.
- 0 OCUsNo compute to size or keep awake. Billing is per GB stored, per GB written and per request.
- $10a month for a small app. Amazon OpenSearch Serverless bills about $350 a month for 2 search OCUs while a collection is awake.
- 2 to 6 msmedian keyword search on 2.8M to 10M documents, measured on 2 to 4 CPUs. AWS publishes no query latency for OpenSearch Serverless.
Your OpenSearch client, a new host
# pip install opensearch-py from opensearchpy import OpenSearch client = OpenSearch( hosts=[{"host": "api.warpsearch.io", "port": 443, "url_prefix": "compat/opensearch"}], use_ssl=True, headers={"Authorization": "Bearer ws_..."}, ) client.indices.create(index="products", body=mapping) client.bulk(body=actions, refresh=True) client.search(index="products", body={"query": {"match": {"title": "wireless headphones"}}})
Up and running in three steps
Create an API key
Sign up and create a key in the console. Keys have roles: viewer searches, writer also indexes, editor also creates indices, admin also deletes them. Send the key as
Authorization: Bearer,ApiKeyorX-API-Key.Create your index with a mapping
PUT /productswith your mappings and analysis settings, as on OpenSearch. There is no dynamic mapping, so send the mapping first or name fields with type suffixes. See Schemas and fields.Point your client at the endpoint
Base URL
https://api.warpsearch.io/compat/opensearch, port 443. Bulk load with_bulk, call_refreshonce at the end, and search. Free up to 1 GB with no card.
Works with the OpenSearch tools you run
Every official client takes a base path and a header. Settings for each are in the connection table.
What serverless means here
Pay per request, not per OCU-hour
$2.00 per million requests, $0.08 per GB stored a month and $1.00 per GB written. A quiet app pays the $10 minimum. There is no 2 OCU floor and no OCU math.
Nothing to size
No OCUs, collections to right-size or capacity limits to set. A collection grows with your data and your traffic, and an idle one costs only its storage.
k-NN and hybrid in one index
knn_vector fields up to 8192 dimensions, the knn query with filters applied inside the search, and the hybrid query with an RRF search_pipeline in the request body. See Vector search.
Logs and search in one service
Time-series data indexes the same way as search data, with storage from $0.08 down to $0.035 per GB a month at volume. Ship with Fluent Bit, query from Grafana. See Log analytics.
99.9% uptime SLA from $10
Every paid account has it, with no replicas or higher tier to buy.
On AWS, or in your AWS account
Pick an AWS, Google Cloud or Azure region close to your app. For data that must stay in your account, WarpSearch runs inside your own AWS account with the same APIs. Ask about BYOC.
How fast is it next to Amazon OpenSearch Serverless?
AWS publishes no query latency, vector recall or indexing rate for OpenSearch Serverless. It does publish how long things wait. Here is what we measured, on deliberately small hardware.
| WarpSearch | OpenSearch Serverless | |
|---|---|---|
| Keyword search | 1 to 6 ms median (2.1 ms on 2.8M docs, 2.4 ms on 10M passages) | Not published |
| Vector search (k-NN) | 6.2 ms median at 0.964 recall@10 on 10M vectors. 10.4 ms with filters | Not published |
| Log filters, 22M events | 1 to 2 ms median | Not published |
| Ingest | 63,000 to 97,000 short documents a second | Not published per OCU |
| Time to searchable | Immediate with refresh=true or _refresh. Otherwise within a few minutes | About 10 seconds, not configurable |
| Idle | Storage only. A collection idle for 15 minutes reloads on its next request; optional keep-warm keeps it ready. We have not published cold-start timings yet | NextGen: 10 to 30 seconds to wake. Classic never sleeps: 2 OCUs minimum |
Measured October 2026 through our Elasticsearch-compatible API, warm cache, client in the same region, single runs. Every test ran on deliberately minimal hardware, 2 to 4 CPUs (a 4 CPU container or an AWS i8g.large with 2 vCPUs), so these are the least favorable results we can produce. More hardware only improves them. Numbers exclude internet round trips.
What serverless OpenSearch costs for two workloads
Monthly cost at list prices in US East for two common search workloads. Open a platform for the line items and sources.
| Platform | Small RAG | Catalog |
|---|---|---|
| WarpSearch | $10 | $149 |
| OpenSearch Serverless | $357 to $53036 to 53x cheaper | $550 to $1,0803.7 to 7.2x cheaper |
| OpenSearch Service | $132 to $39713 to 40x cheaper | $400 to $1,5752.7 to 11x cheaper |
| Aiven | $365 to $73037 to 73x cheaper | $730 to $1,4554.9 to 9.8x cheaper |
| Elastic Serverless | About $100 to $21010 to 21x cheaper | $210+1.4x cheaper or more |
| Elastic Cloud | $93 to $2859.3 to 29x cheaper | $516 to $1,5003.5 to 10x cheaper |
| Pinecone | $118 to $13712 to 14x cheaper | $8,100 to $9,20054 to 62x cheaper |
| turbopuffer | $161.6x cheaper | $536 to $5563.6 to 3.7x cheaper |
| Small RAG | 1M text chunks with 768-dim embeddings, 7 GB of JSON. 1M queries a month spread through the day. 1 GB written a month. |
|---|---|
| Catalog | 5M products with 384-dim embeddings, 10 GB of JSON. 50M queries a month (about 19 a second on average). 20 GB written a month. |
List prices in US dollars, checked October 20, 2026. Ranges cover regions, plan choices, or how the vendor sizes compute. All comparisons and sources.
What serverless OpenSearch includes
This is what the service includes today. A request for anything else answers 404 or 400 unsupported_feature.
| Area | Status | Notes |
|---|---|---|
Search: _search, _msearch, _count, Query DSL | Available | match, multi_match, bool, term, range, prefix, wildcard, regexp, fuzzy, function_score, query_string, geo queries, sort, collapse, post_filter. Scores differ from OpenSearch |
| Aggregations | Most | terms, multi_terms, histograms, ranges, filters, metrics, cardinality, percentiles, top_hits, geo grids. No pipeline aggregations, composite or scripts |
| k-NN and hybrid | Available | knn_vector, knn query with filter, hybrid with an inline RRF pipeline. No neural or neural_sparse queries, no radial search |
| Highlighting | Available | Standard highlight object on search and multi-search |
| Index and mapping APIs | Available | Create with mappings and analysis settings, get, delete, add fields. No dynamic mapping, index templates or aliases |
Documents and _bulk | Most | index, create, delete, doc_as_upsert, get, mget, refresh, forcemerge. No partial updates, update_by_query, delete_by_query or reindex |
| Clients and shippers | Available | All official clients, Fluent Bit, Grafana. Data Prepper is not supported |
| Authentication | API keys | Bearer, ApiKey or X-API-Key with per-key roles. No SigV4, IAM or data access policies, no HTTP Basic |
Scroll, search_after, point in time | Not supported | Page with from and size, or a range filter on the last hit's sort values |
| Nested and join fields, scripts | Not supported | |
| OpenSearch Dashboards, ISM, alerting, snapshots, ML Commons, PPL and SQL plugins | Not supported | Grafana works for dashboards. SQL is on the WarpSearch API |
Full tables: OpenSearch reference.
Questions
What is serverless OpenSearch?
OpenSearch search with no cluster or compute units to size. On WarpSearch you create a collection, point an OpenSearch client at it and pay per GB stored, per GB written and per request, from $10 a month. Amazon OpenSearch Serverless is AWS's version and bills compute in OCU-hours.
How is this different from Amazon OpenSearch Serverless?
Amazon OpenSearch Serverless bills OCU-hours at $0.24 each and needs 2 search OCUs while a collection is awake, about $350 a month before storage, and writes take about 10 seconds to become searchable. WarpSearch has no OCUs, bills each request, and makes writes searchable immediately when you ask with refresh. WarpSearch does not include OpenSearch Dashboards or SigV4 authentication; the table above lists the scope. See the line-by-line cost comparison.
Does it run on AWS?
Yes. WarpSearch runs on AWS, Google Cloud and Azure. It can also run inside your own AWS account with the same APIs. Contact us about BYOC.
Which OpenSearch clients work?
opensearch-py and the JavaScript, Java, Go, .NET, Ruby and PHP clients, with the base URL and an API key header. Fluent Bit ships logs through its es output, and Grafana's OpenSearch data source reads them. Settings for each are in the OpenSearch reference.
Does serverless OpenSearch on WarpSearch support k-NN vector search?
Yes. Map a knn_vector field with up to 8192 dimensions, send the OpenSearch knn query with or without a filter, and combine it with keyword search in a hybrid query with an RRF pipeline in the request body. Vectors live in the same index as your other fields. See Vector search.
Can I use OpenSearch Dashboards or SigV4?
No. WarpSearch serves the OpenSearch REST API, not Dashboards, and authenticates with API keys instead of IAM and SigV4. Grafana's OpenSearch data source works for dashboards. Scroll, search_after, nested documents and scripts are also not supported; the table above has the full scope.
Is this managed OpenSearch?
It is serverless OpenSearch rather than a managed cluster: there are no nodes, instance types, EBS volumes or shards. A 3-node Amazon OpenSearch Service domain is about $390 a month before storage; WarpSearch starts at $10. See the managed OpenSearch comparison.
Read next
Free up to 1 GB, no card. Pay per GB and per request after that.