Algolia vs Elasticsearch
This is the least like-for-like pairing on the site, which is exactly why people search for it. Algolia sells a finished search product priced per request; Elasticsearch sells an engine that will do search, logging, analytics and vector retrieval on one cluster, priced by compute and storage, and it expects you to bring expertise. The axis here is not price per query, it is how much of the search problem you want to own.
Pick Elasticsearch if search is one of several jobs (logs, analytics, vectors) you want on one platform, if you already have people who know clusters, or if you need an OSI-approved licence, since AGPL-3.0 has been an option since September 2024. Stay on Algolia if search is a product feature rather than a platform, and you would rather pay per request than staff mappings, analyzers and shard sizing.
Side by side
Algolia vs Elasticsearch at a glance
| Algolia | Elasticsearch | |
|---|---|---|
| Billing unit | ✗ Per search request, one per keystroke and index by default | ✓ Per VCU-hour plus GB retained, or your own servers |
| Serverless rates | $0.50 per 1,000 requests on Grow, $1.75 on Grow Plus | $0.09 per VCU-hour search, $0.14 ingest, $0.047 per GB retained |
| Worked example | 10,000 requests free, then usage-based with no ceiling | $24 to $27/month for a 2 GB dev environment, per Elastic |
| Free option | ✓ Build: 10,000 requests and 1,000,000 records a month | ✓ Free to self-host, unlimited scale, cloud trial available |
| Licence | ✗ Proprietary SaaS | ✓ AGPL-3.0, SSPL or Elastic License, your choice since 2024 |
| Time to first useful search | ✓ Hours: upload records, ship the widget | ✗ Days to weeks: mappings, analyzers, relevance tuning |
| As-you-type UX | ✓ Built in, typo tolerance by default | ✗ Available, but you build and tune it (edge n-grams, fuzziness) |
| Merchandising for non-engineers | ✓ Rules, banners, A/B tests, personalization in a console | ✗ Query-time logic written by developers |
| Scope beyond site search | ✗ Search and discovery only | ✓ Logs, metrics, security analytics and vector search too |
| Scale ceiling | High, but priced per request all the way up | ✓ Billions of documents on hardware you control |
| Ops burden | ✓ None | ✗ Real: shards, replicas, upgrades, capacity planning |
| Best for | Product teams who want search to be somebody else's job | Platform teams consolidating search and observability |
Sources: Algolia pricing · Elasticsearch Serverless pricing · Elastic licensing FAQ. Serverless prices took effect 1 December 2024. Compiled August 2026.
Where Elasticsearch wins
Cost stops tracking user behaviour
Elastic bills compute and storage: $0.09 per VCU-hour for search, $0.14 for ingest, $0.047 per GB retained monthly on serverless, and nothing but servers if you self-host. Nobody's typing speed appears on the invoice. Elastic's own published examples are $24 to $27 a month for a 2 GB development environment and $190 to $210 for a 20 GB production environment; treat those as vendor illustrations rather than quotes, because real cost depends on retention and query load. The structural point holds either way: this meter counts resources, not keystrokes.
One cluster for search, logs and vectors
If your organisation already runs Elasticsearch for logs or observability, adding product search to it is an incremental cost and an incremental skill, not a new vendor, a new contract and a new SDK. That consolidation argument is the most common honest reason to choose Elasticsearch over a dedicated search API, and it is also why the comparison is unfair in both directions: nobody adopts Elasticsearch to make a docs search box faster.
An OSI-approved licence, since September 2024
Elastic added AGPL-3.0 alongside SSPL and the Elastic License in September 2024, which put Elasticsearch back inside most "must be OSI-approved" procurement policies. AGPL is strong copyleft and its obligations reach network use of modified versions, so it is a heavier commitment than Apache-2.0. If that matters, OpenSearch is the Apache-2.0 fork of the same lineage and has been under Linux Foundation governance since September 2024.
There is no ceiling
Billions of documents, petabyte indices, custom scoring, aggregations that behave like analytics queries: Elasticsearch does the things a hosted search API will eventually tell you are out of scope. Buying that headroom before you need it is a mistake, but knowing where the ceiling is not is worth something on a long-lived system.
Where Algolia still wins
Everything Algolia charges for is the work Elasticsearch hands back to you. Typo tolerance, prefix matching, faceting, ranking that feels right on the first day: on Elasticsearch these are analyzer, mapping and query design decisions, and getting them wrong is invisible until conversion drops. Running Elasticsearch means understanding the JVM, Lucene, shards, replicas, mappings and analyzers, and that knowledge has to exist inside the team permanently, not just during the migration. For a team of five shipping a product, a per-request bill is often cheaper than the engineer-weeks.
Algolia's model also wins outright in the low-traffic case. The free Build plan carries 10,000 search requests and 1,000,000 records a month with no overage, which is more free capacity than any managed Elasticsearch offering, and the merchandising console remains something Elastic does not sell to non-engineers at all. If the search box is a feature rather than a platform, moving to Elasticsearch to save money frequently costs more.
Migrating Algolia to Elasticsearch
The path most teams take:
1. Inventory first. Write down searchable attributes and their priority, custom ranking, synonyms, facets and rules. On Elasticsearch each of these becomes a mapping or query decision, so this document is the design, not paperwork.
2. Design the index before importing. Elasticsearch is explicit where Algolia is schemaless: choose analyzers, decide how prefix and typo matching are implemented, and only then bulk-import. Importing first and tuning later is the classic way to end up reindexing everything.
3. Rebuild the query layer, not just the client. There is no InstantSearch adapter here in the way Typesense and Meilisearch provide one, so expect to write the search API layer and adapt or replace the front-end widgets.
4. Run both for one cycle. Compare relevance on real queries with both engines live, then cut over. Algolia's analytics history, A/B tests and personalization profiles do not come with you, so export what you need for reporting before you cancel.
Common questions
FAQ: Algolia vs Elasticsearch
Is Elasticsearch cheaper than Algolia?
In infrastructure terms it usually is at scale, but it moves cost from an invoice to a team. Elastic Cloud Serverless bills 0.09 dollars per VCU-hour for search, 0.14 for ingest and 0.047 per GB retained each month, and Elastic's own examples put a 2 GB development environment at 24 to 27 dollars a month and a 20 GB production environment at 190 to 210 dollars. Self-hosting costs only servers. What neither figure includes is the engineer time to design mappings, tune analyzers and keep the cluster healthy, which is exactly the work Algolia's per-request price buys you out of.
What licence is Elasticsearch under in 2026?
Elasticsearch is triple-licensed. In September 2024 Elastic added AGPL-3.0, an OSI-approved licence, alongside the Server Side Public License and the Elastic License, so you can take the AGPL option if your policy requires an OSI-approved licence. AGPL obligations extend to network use of modified versions, which matters if you fork the engine and expose it as a service. If you want a permissive licence with no copyleft at all, OpenSearch is Apache-2.0 and forked from Elasticsearch, so the query surface will feel familiar.
Should I choose Elasticsearch or OpenSearch to replace Algolia?
Choose on licence and support rather than on features, because the two are close relatives. OpenSearch is Apache-2.0 and has been under Linux Foundation governance since September 2024, which is the simplest answer for procurement and carries no copyleft. Elasticsearch has the larger ecosystem, more mature machine learning and vector features, and a serverless product with published VCU pricing. If you only need site or product search rather than a logging and observability platform, both are heavier than Typesense or Meilisearch, and the operational cost is the reason to think twice.