If you want to keep CQL but stop running nodes, look at Amazon Keyspaces on AWS or the Azure Cosmos DB API for Apache Cassandra, and check which CQL features they support. If you want a Cassandra-compatible engine with vendor-run operations, ScyllaDB is the closest match, but it is no longer open source. If you are free to change the API, Amazon DynamoDB and Google Cloud Bigtable are managed wide-column and key-value services on their clouds, and Apache HBase fits estates already on Hadoop. If you need SQL and multi-row ACID transactions, YugabyteDB offers both a Cassandra-style API and a PostgreSQL-compatible one under Apache 2.0.
Apache Cassandra is an Apache 2.0, masterless wide-column database queried with CQL. Every node accepts reads and writes, replicas are placed per data centre, and each request picks its own consistency level. 5.0.9 (7 August 2026) is the latest release, and the 4.1 and 4.0 lines are still maintained.
Cassandra is free and proven for write-heavy, multi-data-centre workloads. Teams that look elsewhere usually say one of these; the first three follow from the project's own documentation, the last is our editorial reading.
- Operations. Repair, compaction and capacity planning need specialist skills, and the project offers no managed service of its own; managed options come from third parties such as DataStax Astra DB (now part of IBM) and Amazon Keyspaces.
- Transactions. Cassandra 5.0 has lightweight, single-partition transactions only; there are no multi-partition ACID transactions.
- Query flexibility. CQL has no joins and expects filters on the primary key or an index, so tables are designed per query.
- Cloud alignment. Teams on one cloud often prefer that provider's serverless or managed database to running their own ring (editorial).
Head-to-head pages: Cassandra vs ScyllaDB (planned), Cassandra vs MongoDB and DynamoDB vs Cosmos DB (planned). For sensor data, see Cassandra for telemetry; if you are not sure you need a wide-column store, see SQL vs NoSQL.
Quick picks
Serverless, Cassandra-compatible tables with no nodes to manage.
Runs CQL applications and offers a DynamoDB-compatible API; source-available with a managed X Cloud service.
Key-value and document tables billed per request or capacity unit, with global tables.
Managed wide-column store for very large single-keyed datasets, with an HBase-compatible client.
Apache 2.0 distributed database with a Cassandra-style YCQL API and PostgreSQL-compatible YSQL.
How we chose
We considered ScyllaDB, Amazon Keyspaces, Azure Cosmos DB for Apache Cassandra, DataStax Astra DB, Amazon DynamoDB, Google Cloud Bigtable, Apache HBase, YugabyteDB, Aerospike and MongoDB. To be included, a product had to handle large, write-heavy, partitioned workloads like Cassandra's, be actively developed or sold in November 2026, and document its compatibility or data model and licence on an official page.
Astra DB is mentioned but not listed separately, because we did not re-check its current terms after the IBM acquisition. Aerospike (a key-value store) and MongoDB (a document store) solve neighbouring problems; see their pages and Cassandra vs MongoDB. ScyllaDB is included with a clear note: its open source edition ended with 6.2, and current releases are source-available with free use limited to organisations that are not paying customers, up to 10 TB and 50 vCPUs.
The order is not a ranking. Cassandra-compatible options come first, then other wide-column and key-value services, then distributed SQL.
At a glance
| Option | CQL compatible | Licence or model | Where it runs | Transactions | Main trade-off |
|---|---|---|---|---|---|
| ScyllaDB | Yes (plus DynamoDB API) | Source-available; free for non-customers to 10 TB and 50 vCPUs | Self-managed or ScyllaDB X Cloud | Lightweight transactions | Not open source; no published prices |
| Amazon Keyspaces | Yes (CQL 3.11 API), with gaps | Managed AWS service | AWS only | Not reviewed here | No secondary indexes, UDFs or materialized views |
| Azure Cosmos DB for Apache Cassandra | Yes (Cassandra drivers) | Managed Azure service, RU billing | Azure only | Not reviewed here | Request-unit pricing to model |
| Amazon DynamoDB | No (own API, PartiQL subset) | Managed AWS service | AWS only | ACID, up to 100 actions | Rewrite data access; 400 KB items |
| Google Cloud Bigtable | No (client libraries, HBase API) | Managed Google Cloud service | Google Cloud only | Single-row atomic only | Paid node-hours around the clock |
| Apache HBase | No (Java API; SQL via Phoenix) | Apache 2.0 | Self-managed on Hadoop | Single-row atomic only | Needs HDFS and ZooKeeper |
| YugabyteDB | YCQL is rooted in CQL | Apache 2.0 core; Aeon managed | Self-managed or Aeon on three clouds | Distributed ACID | Different engine; plan a migration, not a swap |
ScyllaDB
ScyllaDB is a wide-column database written in C++ that speaks CQL, so Cassandra drivers and data models carry over, and its Alternator interface is compatible with the DynamoDB API. Recent releases use tablets for data distribution (recommended for new clusters since 2026.1) and Raft for cluster metadata; 2026.3 (September 2026) added full-text search with BM25 scoring and cluster-wide restore. ScyllaDB X Cloud is the vendor-run managed service.
Read the licence before you plan on it. ScyllaDB Open Source 6.2.x (AGPL) was the final open source release; from 2025.1 there is one source-available stream, and the current licence (v1.1, April 2026) is free only for organisations with no paid relationship in the previous 12 months, up to 10 TB of storage and 50 vCPUs in total, and forbids offering it as a service. See Cassandra vs ScyllaDB.
- No longer open source; most production estates need a quoted commercial licence or X Cloud.
- The licence assigns modifications to ScyllaDB and allows audits, licence keys and telemetry.
- No published prices for X Cloud or self-managed Enterprise.
Amazon Keyspaces
Amazon Keyspaces (for Apache Cassandra) is a serverless, Cassandra-compatible database service: there are no servers to provision or patch, tables scale with traffic, and AWS says you can use the same Cassandra application code and developer tools. It is compatible with the CQL 3.11 API (backward-compatible with 2.x), supports logged and unlogged batches, TTL, static columns and JSON in SELECT and INSERT, and replaces Cassandra's security settings with IAM.
Its supported-API list also shows what is missing: CREATE INDEX, user-defined functions and aggregates, triggers, materialized views, TRUNCATE and ALTER TYPE are not supported. Compaction, repair and similar node settings are not applicable because AWS runs the storage. AWS publishes a calculator that compares Keyspaces costs with an existing cluster's nodetool status output.
- No secondary indexes, UDFs, aggregates, triggers or materialized views.
- AWS only.
- Costs follow request volume and storage, so model them with your own traffic.
Azure Cosmos DB for Apache Cassandra
Azure Cosmos DB offers an API for Apache Cassandra, so applications use Cassandra drivers against a managed, globally distributed service. Like the other Cosmos DB request-unit APIs, it is billed for provisioned throughput (standard or autoscale) or for request units consumed (serverless), plus storage in every region an account is replicated to. The lifetime free tier gives one account per subscription 1,000 RU/s and 25 GB.
We have not reviewed its CQL feature coverage in detail, so check Microsoft's compatibility documentation against your schema and queries. See Azure Cosmos DB explained.
- Request-unit costs depend on item size, indexing and query shape.
- Feature compatibility with Cassandra needs checking before migration.
- Azure only.
Amazon DynamoDB
DynamoDB is AWS's serverless key-value and document database. Like Cassandra, it is designed around a partition key and optional sort key, so query-first modelling carries over, but the API is different (GetItem, Query, Scan and a subset of PartiQL). It adds ACID transactions of up to 100 actions, Streams for change capture and global tables, with a choice of multi-Region eventual or strong consistency fixed at creation.
On-demand mode bills per request and provisioned mode per capacity unit hour; AWS states a 99.99% SLA for single-Region tables and 99.999% for global tables. ScyllaDB's Alternator API is compatible with DynamoDB, which can matter if you want a way back off AWS later. See DynamoDB vs Cosmos DB.
- Not CQL-compatible, so data access code must be rewritten.
- Items are limited to 400 KB.
- AWS only.
Google Cloud Bigtable
Bigtable is a managed wide-column store: rows have a row key, columns are grouped into column families, and each cell can keep timestamped versions. Storage sits on Google's Colossus file system, so nodes provide compute and can be added or removed without moving data. A single-cluster instance is strongly consistent; replication across clusters is eventually consistent by default and configurable per app profile. Since April 2026 it has Enterprise and Enterprise Plus editions.
Applications use Google's client libraries, including an extension of the HBase Java client, and GoogleSQL for SELECT queries only. You pay for provisioned node-hours whether or not they are busy, plus storage per replica. See time series databases for related options.
- Writes are atomic per row only; there are no multi-row transactions.
- SQL is read-only (no DML, joins or subqueries).
- Google Cloud only, with nodes billed around the clock.
Apache HBase
HBase is the Hadoop project's Bigtable-style wide-column store under Apache 2.0. Tables are split into regions by row-key range and served by RegionServers, data lives in HDFS, and reads and writes are strictly consistent per row, which is a different trade-off from Cassandra's tunable consistency. Apache Phoenix adds SQL and secondary indexes on top. The downloads page lists 2.5.16 as the stable release, with 2.6.7 and 3.0.0 also published.
It makes sense where HDFS, ZooKeeper and Hadoop skills already exist; outside that world most new projects pick a lighter or managed option (editorial).
- Depends on HDFS and a ZooKeeper quorum.
- No SQL of its own and no multi-row ACID transactions.
- Each region is served by one RegionServer at a time, a different availability model from Cassandra's masterless replicas (editorial).
YugabyteDB
YugabyteDB offers two APIs on one Raft-replicated storage layer: YCQL, a semi-relational API rooted in the Cassandra Query Language, and YSQL, which reuses the PostgreSQL query layer. Both support distributed ACID transactions, and the FAQ describes the system as CP, with writes acknowledged by a majority of a tablet's replicas. The database is Apache 2.0, and YugabyteDB Aeon is the managed service on three clouds, with a free single-node Sandbox.
It suits teams that chose Cassandra for scale and resilience but now need transactions or relational queries. YCQL is rooted in CQL rather than identical to it, so treat the move as a migration. See CockroachDB alternatives for distributed SQL options.
- Strong consistency through Raft means a different latency and availability trade-off from Cassandra's tunable consistency.
- YCQL is not a drop-in Cassandra replacement.
- YugabyteDB Anywhere and Aeon Enterprise are priced by quote.
When to stay with Cassandra
Stay with Cassandra if you value its Apache 2.0 licence with no usage limits, masterless writes that continue through node and data-centre failures, and a choice of managed providers. Cassandra 5.0 added Storage Attached Indexes, the Unified Compaction Strategy and a vector type, which address some reasons teams used to leave. If operations are the pain, a managed Cassandra-compatible service is a smaller step than a new data model.
Switching makes most sense when you want a cloud provider to run it (Keyspaces, Cosmos DB, DynamoDB, Bigtable), when you already run Hadoop (HBase), or when you need multi-row transactions and SQL (YugabyteDB). ScyllaDB is the closest drop-in engine, but weigh its licence terms against Cassandra's.
Frequently asked questions
What is the closest alternative to Cassandra?
For CQL compatibility, ScyllaDB (source-available) and the managed services Amazon Keyspaces and the Azure Cosmos DB API for Apache Cassandra. Each documents its own compatibility limits, so test your schema and queries.
Is ScyllaDB open source?
Not any more. ScyllaDB Open Source 6.2.x (AGPL) was the final open source release; from 2025.1 ScyllaDB is source-available, free only for non-customers up to 10 TB and 50 vCPUs.
Does Amazon Keyspaces support all Cassandra features?
No. It is compatible with the CQL 3.11 API, but AWS lists secondary indexes, user-defined functions and aggregates, triggers, materialized views and TRUNCATE as not supported.
Which Cassandra alternative supports ACID transactions?
YugabyteDB supports distributed ACID transactions through both its YCQL and YSQL APIs, and DynamoDB supports ACID transactions of up to 100 actions. Cassandra 5.0 has lightweight, single-partition transactions only.
Is MongoDB an alternative to Cassandra?
For some workloads. MongoDB is a document database with richer ad hoc queries; Cassandra is built for write-heavy, multi-data-centre tables. See Cassandra vs MongoDB and MongoDB alternatives.
Sources
- Apache Cassandra downloads
- Apache Cassandra 5.0 new features
- ScyllaDB source-available FAQ
- ScyllaDB licence file
- What is Amazon Keyspaces
- Amazon Keyspaces: supported Cassandra APIs
- Microsoft Learn: Azure Cosmos DB introduction
- Amazon DynamoDB pricing
- Bigtable editions overview
- Apache HBase downloads
- YugabyteDB GitHub repository
- YugabyteDB pricing
Checked November 2026.
How we research these guides: our editorial method.