Azure Cosmos DB

Azure Cosmos DB is Microsoft's managed database service in Azure. Microsoft calls it a fully managed NoSQL and vector database and says it stores document, vector, key-value, graph and table data. Several APIs sit on top of one service: NoSQL, MongoDB, Apache Cassandra, Apache Gremlin and Table, with PostgreSQL and a MongoDB vCore option on the pricing page.

Microsoft says the service handles patching and capacity so that your team does not run database servers, and that data can be copied to any Azure region and written to in several regions at once. The documentation lists single-digit millisecond response times and a 99.999 percent availability SLA for multi-region writes as its headline claims. Microsoft also says it is a poor fit for heavy analytics, which it sends to Microsoft Fabric, and for highly relational applications, which it sends to Azure SQL.

• Global distribution. You add or remove Azure regions on an existing account without pausing the application. With multi-region writes every region accepts reads and writes, and Microsoft quotes a 99.999 percent SLA for that setup.
• Five consistency levels. Strong, bounded staleness, session, consistent prefix and eventual. Microsoft says most distributed NoSQL databases offer only strong and eventual, and that each level trades availability and speed differently.
• Vector and hybrid search. Embeddings can be stored beside operational data and searched with DiskANN-based similarity, with hybrid search also listed. Microsoft positions this for retrieval-augmented generation, AI agents and caching of LLM answers.
• Change feed. A log of changes to a container that can feed Azure Functions, Spark or streaming pipelines. Microsoft lists latest-version and all-versions-and-deletes modes for event-driven designs.
• Analytics without ETL. Azure Synapse Link and Fabric mirroring give an isolated analytical store, so reporting queries do not compete with the transactional workload for throughput.
• Continuous backup. Point-in-time restore is available with a retention window of 7, 30 or 35 days, depending on the tier chosen. Backups are written to Azure Blob Storage in each write and read region.
• Hierarchical partition keys. Sub-partitioning lets multitenant and high-cardinality workloads grow beyond the limit of a single logical partition, according to the product documentation.

• A retailer with shoppers in three continents. Microsoft lists retail and personalisation as target workloads. A team can place replicas in regions near customers, keep session consistency for a shopping cart, and add a region before a sales peak without redeploying the application.
• A software house building an AI agent product. Conversation state, documents and embeddings can sit in one account. Microsoft's introduction says it provides the data layer for AI agents, and quotes Satya Nadella saying OpenAI relies on Cosmos DB to scale ChatGPT.
• An IoT team collecting device state and logs. Microsoft names IoT device state and logs as a high-throughput fit. The service integrates with IoT Hub and Functions, and autoscale throughput adjusts between 10 and 100 percent of a set limit as device traffic changes.

Sign in to the Azure portal with a subscription and create an Azure Cosmos DB account, choosing an API first. Microsoft states that once a compute pricing model and API are chosen, they cannot be changed, so the decision needs care. During creation you can opt in to the free tier.

Then you create a database and container, pick a partition key, and set throughput as standard, autoscale or serverless. SDKs are listed for .NET, Java, JavaScript and Node.js, Python, Go and Rust. A local emulator is available for development at no cost. The vendor offers a capacity planner calculator to estimate request units before you commit.

Cosmos DB offers an API for MongoDB over the wire protocol. Microsoft's own page directs migrations that need full MongoDB Query Language to Azure DocumentDB. The free tier is one per account.

Azure Cosmos DB is entry #18 of 22 in the database management systems list: it follows Amazon DocumentDB and comes before SurrealDB. From place 4 the established names follow in their ranked order; we publish no numeric score.

Azure Cosmos DB is based in the United States; 15 of the 22 products here share its region, North America.

Pricing as recorded: Request units per second, autoscale or standard, or serverless; free tier of 1,000 RU/s and 25 GB per month.

Microsoft says it bills for compute, storage and bandwidth. Compute is measured in request units per second (RU/s) under standard provisioned, autoscale provisioned or serverless models, or per vCore node for the PostgreSQL and MongoDB vCore options. Autoscale starts at a 1,000 RU/s limit and bills on the highest RU/s used each hour between 10 and 100 percent of that limit. The free tier gives the first 1,000 RU/s and 25 GB per account, one account per subscription. Reserved capacity gives 20 percent off for one year and 30 percent for three years on the autoscale page. The per-unit prices did not appear in the page text we fetched.

Checked against azure.microsoft.com/en-us/products/cosmos-db for Best Database Management Systems in 2026, review date 7 November 2026.

The MongoDB introduction lists wire-protocol support, automatic sharding, tunable consistency with five levels and Synapse Link.

It suggests Azure DocumentDB for MQL-heavy migrations.

Data can live in any Azure region, and the pricing page lists regions in Europe including West Europe, North Europe, Germany, France, Sweden and Switzerland. Serverless accounts run in a single region only.

Microsoft states that encryption at rest is available with customer-managed keys, access is controlled by Azure role-based access control, and the Azure platform holds more than 100 compliance certifications. Which certifications apply to Cosmos DB itself was not listed on the pages we read. Export options were not described there either.

Integrations the vendor names (12):

• Azure Functions
• Azure IoT Hub
• Azure Kubernetes Service
• Azure App Service
• Azure Synapse Link
• Microsoft Fabric mirroring
• Apache Spark
• MongoDB drivers
• Apache Cassandra drivers
• Apache Gremlin
• Azure Table Storage
• Azure Cosmos DB emulator

Also in our records:

• Ecosystem tags in our records: Multiple APIs.

We pick the closest products in the same guide by tags, price model and region.

• MongoDB (#5 of 22): Document database sold mainly as Atlas cloud, with a source-available server. Suits: Teams building around JSON documents who prefer a managed cloud service. (United States)
• ScyllaDB (#15 of 22): Wide-column database compatible with Cassandra and DynamoDB APIs. Watch out: wide-column model without joins or flexible document queries. (United States)
• SurrealDB (#19 of 22): Multi-model database under BSL 1.1 with a free cloud instance. (United Kingdom)

Read from the vendor’s own pages on 5 November 2026.

• azure.microsoft.com/en-us/products/cosmos-db: Product overview, security and compliance claim
• learn.microsoft.com/en-us/azure/cosmos-db/introduction: Feature list, free options, good and poor fit workloads, SDK languages
• azure.microsoft.com/en-us/pricing/details/cosmos-db/autoscale-provisioned: Pricing model, autoscale rules, reserved capacity, API list
• learn.microsoft.com/en-us/azure/cosmos-db/free-tier: Free tier limits
• learn.microsoft.com/en-us/azure/cosmos-db/consistency-levels: Five consistency levels
• learn.microsoft.com/en-us/azure/cosmos-db/continuous-backup-restore-introduction: Restore windows of 7, 30 and 35 days
• learn.microsoft.com/en-us/azure/cosmos-db/distribute-data-globally: Multi-region writes, serverless single-region limit

Four tests decide a position: how soon the product is useful, what the bill really becomes, how data leaves and who it suits. The result for Azure Cosmos DB: #18, out of 22 ranked. How vendor facts are sourced, and how money can and cannot influence a page, is in the editorial process.