OpenSearch Pricing Explained (2026): Every Deployment Model, Every Hidden Cost
OpenSearch carries no license fee, and that fact hides one of the widest cost ranges in cloud infrastructure. The same engine runs on a low-cost single node from a budget cloud provider like DigitalOcean and also powers three-Availability-Zone (AZ) Amazon Web Services (AWS) production domains that run more than $1,600 a month.
Your final bill depends on who hosts the cluster and which charges never appear on that host’s main pricing page. Three hosting models cover the market, including AWS managed (provisioned or serverless), third-party managed, and self-hosted.
This guide covers what a provisioned AWS domain bills across instances and Amazon Elastic Block Store (EBS), where the serverless compute-unit floor bites, and how DigitalOcean, Oracle Cloud Infrastructure (OCI), and self-hosted infrastructure compare at the same three workload sizes. All rates quoted here are US East (N. Virginia) on-demand rates unless noted otherwise.
OpenSearch Pricing at a Glance
Entry cost spans more than an order of magnitude across hosts running the same engine. The table shows the starting cost for each deployment model before you factor in workload storage and query volume.
| Deployment | Representative entry cost |
| AWS provisioned (on-demand) | About $27 per month for one t3.small.search at $0.036 per hour ($26.28) plus the 10 GB gp3 minimum at $0.122 per GB-month ($1.22), single-AZ |
| AWS Serverless Classic | About $350 per month for the two-OpenSearch Compute Unit (OCU) production floor at $0.24 per OCU-hour, before storage. A dev/test collection halves that to about $175 |
| AWS Serverless NextGen collections | No compute charge while idle, because OCUs scale to zero after 10 minutes of inactivity. Hot storage runs $0.30 per GB-month, so 10 GB costs about $3 per month |
| DigitalOcean | $19.60 per month for a single Basic node with 2 GiB of RAM, 1 vCPU, and 40 GiB of storage included |
| OCI | About $119 per month in pass-through infrastructure for the smallest cluster. The management fee is waived for up to two data nodes and runs $0.25 per node-hour after that |
| Instaclustr | $62.36 per month for a single fully provisioned t4g.small on AWS with 2 GiB of RAM and 5 GiB of storage. The same node costs $16.25 under Bring Your Own Cloud, which covers management only |
| Self-hosted | About $18 per month for one t3.small Amazon Elastic Compute Cloud (EC2) instance at $0.0208 per hour ($15.18) plus 40 GB of gp3 at $0.08 per GB-month ($3.20), excluding your operational labor |
*Instaclustr fixes storage to each node size rather than letting you provision it separately, so its entry node carries 5 GiB.
Instance identifiers like t3.small.search and m7g.large.search refer to specific AWS server sizes, where the prefix (e.g., m7g) indicates the hardware family, the middle segment (e.g., large) indicates the amount of compute and memory allocated, and the suffix (e.g., search) marks the instance as an OpenSearch Service variant.
AWS OpenSearch Service Pricing for Provisioned Clusters
Your provisioned Amazon OpenSearch Service domain bills across four meters. You pay some of them continuously and trigger others through storage tiers or delayed upgrades. Review all four before you compare a managed domain with serverless or self-hosted infrastructure.
- Instance-hours: AWS bills compute separately for every node in your domain.
- EBS storage: AWS charges for the provisioned storage attached to every node.
- Warm-tier capacity: You pay for dedicated warm nodes and their managed storage.
- Version support: Older engine versions can trigger automatic surcharges.
The last two meters appear once you tier data or fall behind on upgrades.
What the Free Tier Covers and Excludes
New AWS accounts get 750 hours per month of a t2.small.search instance plus 10 GB of EBS storage for the first 12 months, single-AZ only. Multi-AZ production domains bill at standard rates from day one, and the 10 GB cap holds a demo index rather than a production log stream. After the twelfth month, that same single-AZ domain bills at standard on-demand rates.
On-Demand Rates for Current-Generation Graviton Instances
Representative rates for current-generation Graviton instances are $0.135 per hour for m7g.large.search, $0.178 for r7g.large.search, and $0.120 for c7g.large.search. A worked pricing example puts an m7g.medium.search at $0.068 per hour, or $1.632 per day of continuous operation. Instance-hours bill per node, so a three-node domain multiplies each of those rates by three.
Reserved Instances vs. Database Savings Plans
Reserved Instances (RIs) discount on-demand rates 31 percent for one-year no-upfront and 52 percent for three-year all-upfront. Database Savings Plans added OpenSearch coverage on March 5, 2026, at up to 35 percent for a one-year, no-upfront spend commitment that flexes across families, Regions, and serverless. RIs lock you to one instance type in one Region and are nonrefundable, which makes a Savings Plan the better default for most teams.
EBS Storage Multiplies Across Nodes
EBS gp3 runs $0.122 per GB-month, against $0.135 for older gp2 volumes. AWS charges per provisioned gigabyte per node, so a three-node domain with 500 GB on each node bills 1,500 GB every month. A three-AZ production domain with three r6g.xlarge.search data nodes, three cluster manager nodes, two UltraWarm nodes, and that 1,500 GB on gp2 totals $1,604.83 per month. Moving those volumes to gp3 trims about $19.50 of that.
UltraWarm, Cold Storage, and Managed Storage Tiers
UltraWarm managed storage costs $0.024 per GB-month, roughly 80 percent below hot gp3, but warm data needs dedicated ultrawarm1.medium.search nodes at $0.238 per hour, becomes read-only, and carries no RI option.
Cold storage uses the same rate with no standing node fee, and querying it means migrating back to warm.
The managed storage families keep replicas in Amazon Simple Storage Service (Amazon S3) at that managed-storage rate, so per-node EBS multiplication disappears. A managed-storage example prices 1,500 GB at $36 per month.
AWS OpenSearch Serverless Pricing
Serverless replaces instances with OCUs. Classic collections bill a minimum whether or not you use them, and NextGen collections removed that floor in 2026, which changes the math for anything idle. Indexing and search meter separately, so you must account for both dimensions when estimating that minimum.
How OCU-Hours Work
One OCU bundles roughly 6 GB of random access memory (RAM) with corresponding virtual central processing unit (vCPU) capacity, and it bills at $0.24 per OCU-hour on an hourly basis with per-second granularity.
On Classic collections, that OCU also covers gp3 storage for frequently accessed data and data transfer to managed storage. Indexing and search meter separately at that rate, and Direct Query bills its own OCU type of 2 vCPU and 8 GiB of RAM at the same $0.24. Your first production Classic collection in an account starts at two OCUs, while a dev/test collection starts at one when you disable redundancy.
The exact minimum varies with your data size and collection type.
The Minimum Floor: ~$350 per Month on Classic Collections
Two OCUs across 730 hours works out to roughly $350 per month before a single query or gigabyte, and the one-OCU dev/test setup still costs about $175. AWS made NextGen collections generally available on May 28, 2026, and their OCUs scale to zero after 10 minutes of inactivity, so an idle NextGen collection pays for storage only.
A Classic collection left running after a project ends keeps billing that floor until someone deletes it.
The Bedrock Knowledge Base Cost Trap
Amazon Bedrock’s Quick create option provisions an OpenSearch Serverless vector search collection behind your Knowledge Base automatically, and it meters OCUs from the moment that collection exists. Deleting the Knowledge Base leaves the collection in place.
You have to delete the vector store yourself, and until you do, the Classic minimum keeps billing. Delete the Knowledge Base and its data source first, then open Serverless Collections in the OpenSearch console and delete the collection Bedrock created.
Serverless Storage Bills Separately from OCUs
You pay $0.024 per GB-month to store data in Classic collections, with no per-node multiplication, while NextGen collections price hot storage at $0.30 per GB-month. On 100 GB, that is $2.40 a month on Classic against $30 on NextGen collections. At low query volumes, the zero compute floor tends to outweigh the higher storage rate.
OpenSearch Ingestion and Vector Workload Pricing
Your OpenSearch Ingestion pipelines bill at the standard OCU rate, metered per minute, and a paused pipeline consumes nothing, though no Savings Plan covers Ingestion OCUs. Graphics processing unit (GPU)-accelerated vector indexing OCUs carry that rate on different hardware and can deliver up to 75 percent lower indexing OCU cost on write-heavy vector workloads.
Automatic Semantic Enrichment also bills at the standard OCU rate and costs $51.84 to process 10 GB. Amazon S3 Vectors is the storage-side alternative for vector workloads, at $0.06 per GB-month. At 100 million vectors, the same workload costs $538.32 on OpenSearch Serverless against $116.57 on S3 Vectors.
S3 Vectors costs less because it carries no minimum compute charge to spread across the workload, while OpenSearch Serverless bills the two-OCU floor regardless of query volume.
Extended Support and Data Transfer Bill Outside the Pricing Page
Domains on older engine versions pay an automatic surcharge the day Standard Support ends. From November 7, 2026 through November 7, 2027, OpenSearch 1.0 to 1.2 and 2.3 to 2.9 (plus Elasticsearch 7.1 to 7.8 and earlier) pay a charge equal to the full instance cost, so a three-node m7g.medium.search cluster jumps from $148.92 to about $297.84 per month. A separate flat tier of $0.0065 per normalized instance hour applies to a different set of versions from November 7, 2027.
AWS provides free traffic between nodes inside a single domain. The exemption covers cross-AZ rebalancing and movement to UltraWarm, cold, or S3. Internet egress bills at $0.09 per GB after the first 100 GB each month, for the first 10 TB, and ingress is free. Self-managed EC2 clusters get no such exemption and pay cross-AZ rates in both directions.
OpenSearch Pricing Beyond AWS
Outside AWS, four deployment paths cover most OpenSearch workloads. Instaclustr offers a managed service across multiple clouds with a split-billing model that separates management from infrastructure. DigitalOcean runs a straightforward per-node managed offering aimed at small clusters. Oracle Cloud Infrastructure passes through raw infrastructure rates and waives its management fee at small scale.
Self-hosting on your own EC2 instances remains the lowest-cost path once your team can absorb the operational work. Each option shifts a different portion of cost between management, infrastructure, and internal labor.
Instaclustr Managed OpenSearch
Instaclustr splits the bill differently. Its public pricing table lists per-node and per-cluster monthly rates for OpenSearch across AWS, Azure, Google Cloud, and on-premises, filterable by region and node size. Two account types change what those rates buy. Bring Your Own Cloud covers Instaclustr’s management layer alone, and you pay the cloud provider directly for virtual machines, storage, and firewalls. Run In Instaclustr Account bundles both into one figure.
The gap between those two numbers shows what management costs. A t4g.small on AWS in US East (N. Virginia) runs $16.25 per node per month under Bring Your Own Cloud and $62.36 fully provisioned. An r6g.large with 400 GiB runs $200 against $548.91. Management lands near a quarter of the all-in price on the small node and closer to a third on the larger one.
Storage comes fixed to each node size instead of provisioned on its own, so a t4g.small ships as either a 5 GiB or a 30 GiB variant at $62.36 and $82.32 respectively. That removes the per-node EBS multiplication you get on a provisioned AWS domain, and it also removes your ability to add capacity without moving to a larger node. Annual commitments earn tiered discounts, and special-purpose node configurations still route to a sales quote.
DigitalOcean Managed OpenSearch
A single-node DigitalOcean cluster costs $19.60 per month for 2 GiB of RAM, 1 vCPU, and 40 GiB of storage. Storage beyond that allotment costs $0.215 per GiB per month in 10 GiB increments, and larger single nodes run $37.05 for 4 GiB of RAM and $76.25 for 8 GiB. Dedicated three-node clusters start at a higher base price.
DigitalOcean excludes traffic to and from the database from bandwidth allowances, and your compute floor remains the cost of the node itself.
Oracle OCI Search with OpenSearch
OCI passes through standard infrastructure rates and waives its management fee for clusters with up to two data nodes. Starting with the third data node, OCI charges an additional $0.25 per hour for each extra node. That waiver lets you run one- and two-node clusters at infrastructure rates only.
The smallest cluster OCI’s own limits allow, meaning one data node at 1 OCPU and 20 GB of memory with 50 GB of storage, one leader node at the same size, and one dashboard node at 1 OCPU and 8 GB, works out to about $119 per month once you add the load balancer that OCI provisions automatically.
Self-Hosted OpenSearch
OpenSearch uses the Apache 2.0 license, so self-hosting costs what your infrastructure and people cost. A minimal three-node cluster on r6g.large EC2 instances runs about $221 per month in compute at $0.1008 per hour, plus about $120 for 1,500 GB of gp3 volumes at $0.08 per GB-month, which is the raw EC2 rate and sits below the managed-service rate. That specified configuration costs about $341.
Snapshots and cross-AZ transfer cost extra. The estimate also excludes on-call and upgrade work, and self-hosting sidesteps Extended Support surcharges only if you perform the upgrades.
What Three Workload Sizes Cost per Month
Provisioned and self-hosted columns assume three data nodes across three Availability Zones. The serverless column shows the Classic two-OCU floor plus managed storage, and OCU consumption climbs above that floor with load.
The comparison excludes Oracle as an infrastructure pass-through and Instaclustr because its fixed per-node storage tiers do not line up with the provisioned volumes in the other columns.
| Workload | AWS provisioned (on-demand) | AWS Serverless (Classic) | DigitalOcean | Self-hosted (infra only) |
| 50 GB hot | ~$314 (3× m7g.large.search + 150 GB gp3) | ~$352 floor (2 OCUs + 50 GB managed storage) | ~$22 (single 2 GiB node + 10 GiB added storage) | ~$233 (3× r6g.large + 150 GB gp3) |
| 500 GB per node | ~$573 (3× r7g.large.search + 1,500 GB gp3) | ~$386 floor (2 OCUs + 1,500 GB managed storage) | Dedicated cluster plus storage | ~$341 (3× r6g.large + 1,500 GB gp3) |
| 2 TB total | ~$835 (3× r7g.large.search + 500 GB hot gp3 + two ultrawarm1.medium.search + 1.5 TB warm storage) | ~$400 floor (2 OCUs + 2,048 GB managed storage) | Dedicated cluster plus expanded storage | ~$381 (3× r6g.large + 2,000 GB gp3) |
The two dedicated ultrawarm1.medium.search nodes cost $347.48 per month, which exceeds the roughly $151 the warm tier saves on storage at this volume, so tiering only pays at multi-terabyte warm volumes. Reserved Instances pull the compute portion of the AWS columns down by roughly a third under common one-year OpenSearch terms, while commitments do not affect storage lines.
The AWS Pricing Calculator models a provisioned domain line by line against your own instance and storage mix.
Is AWS OpenSearch Worth It?
Managed OpenSearch earns its premium when your data already lives in AWS and Ingestion pipelines feed it. That case strengthens when your Bedrock retrieval-augmented generation (RAG) stack shares one serverless collection across several Knowledge Bases, and your compliance-bound team also inherits the managed service’s certifications.
It stops earning its premium on idle and low-volume work. You can index a small document set and still pay the full Classic minimum. If your team lacks OpenSearch expertise, you may leave orphaned collections billing after a project ends. And if you run OpenSearch solely for log management, you can face substantial annual bills.
How to Reduce Your OpenSearch Bill
You can trace most oversized OpenSearch bills to a handful of defaults left unchanged. Five tactics cover almost all of the recoverable spend. Capacity changes and data tiering usually produce the strongest savings, and timely upgrades avoid support surcharges.
- Right-size from CloudWatch data: Sustained low CPUUtilization with healthy JVMMemoryPressure and OldGenJVMMemoryPressure signals headroom to drop an instance size. SearchLatency confirms whether users would notice. Running each team’s domain in its own account keeps their spend separately attributable.
- Commit to a discount: For steady workloads, evaluate the available commitment discount options against on-demand pricing.
- Tier aging indexes: 10 TB costs about $1,249 per month hot on gp3 and about $246 on UltraWarm, roughly $1,003 in savings before warm-node costs. Index State Management (ISM) automates the hot-warm-cold-delete flow, and Index Rollups aggregate aging data into coarser time buckets.
- Cut shard counts: Oversharded domains burn CPU and heap on coordination, and consolidating small indexes can free enough capacity to remove hot nodes. Inter-node traffic needs no tuning because intra-domain transfer is free.
- Upgrade before Extended Support lands: The surcharge above doubles compute on affected versions. On serverless, cap your maximum OCUs and watch OCUUtilization.
Apply these tactics against measured workload behavior instead of changing every domain at once. Your capacity and retention choices usually provide the clearest path to sustainable savings.
When Does an Observability Platform Make More Sense?
Every OpenSearch billing model you have seen so far charges you for capacity rather than data. Provisioned domains and Classic serverless collections bill for reserved or minimum capacity, and NextGen collections still meter active compute and storage even after removing the idle floor. For teams running OpenSearch as a log analytics backend, that structure means paying for headroom your workload rarely uses.
A per-gigabyte ingestion model inverts that relationship. Coralogix meters observability data at $0.42 per GB for logs, $0.16 per GB for traces, and $0.05 per GB for metrics, and charges nothing when no data arrives. The observability cost optimization guide and the cost optimization page explain how that model works in practice.
OpenSearch remains a solid application search engine, though a per-GB observability platform is often the better fit for log analytics workloads. If your OpenSearch spend skews toward logs, run your actual log volume through a free 14-day Coralogix trial and compare it against the ingestion rates above before you renew a reservation.
Frequently Asked Questions About OpenSearch Pricing
Is AWS OpenSearch the same as Elasticsearch?
No. OpenSearch forked from Elasticsearch 7.10 in 2021 after Elastic moved off the Apache 2.0 license, AWS renamed its service to Amazon OpenSearch Service that September, and the Linux Foundation’s OpenSearch Software Foundation now governs the project. The services remain distinct even though OpenSearch began from that Elasticsearch codebase.
How is OpenSearch OCU pricing calculated?
An OCU bundles roughly 6 GB of RAM with corresponding vCPU capacity and bills at $0.24 per OCU-hour with per-second granularity, charged at the collection group level. Production Classic collections start at the two-OCU minimum described above, dev/test can start with one OCU, and NextGen collections charge only storage while idle.
Does OpenSearch have a pricing calculator?
Yes, the AWS Pricing Calculator models provisioned domains line by line, and it is where to check OR2, OM2, and OI2 rates, which the service pricing page does not reliably publish. It does not model your workload’s query volume or the OCU consumption that follows from it.
What is the cheapest way to run OpenSearch?
OpenSearch itself costs nothing because it uses the Apache 2.0 license, and AWS’s free tier provides a small single-AZ instance plus limited EBS capacity for 12 months. DigitalOcean offers the cheapest managed entry point in this comparison at $19.60 per month, while self-hosting becomes the lowest-cost option at the two TB three-node workload under the table’s infrastructure assumptions. Neither comparison includes your operational labor.
What are the downsides of using OpenSearch?
An idle Classic serverless collection bills its minimum whether or not anything queries it, and a provisioned domain charges every provisioned gigabyte once per node. Old engine versions attract Extended Support surcharges. Self-hosted clusters move upgrades and snapshots onto your team. Your team also handles on-call work.
