Neo4j vs ArangoDB

Quick verdict

Neo4j stores a property graph: nodes with labels and properties, connected by typed relationships that also carry properties. You query with Cypher, a declarative pattern-matching language; since Neo4j 2025.06 new Cypher features go only into Cypher 25, while Cypher 5 is frozen. Neo4j uses calendar versioning: 2026.10.2 (6 November 2026) is current, and the 5.26 LTS line is at 5.26.33. Community Edition is GPL v3; Enterprise Edition adds clustering, online backup, role-based access control and LDAP; AuraDB runs Neo4j on AWS, Google Cloud and Azure. See our Neo4j profile.

ArangoDB stores JSON documents in collections, and edge collections link documents into graphs, so the same data can be read as documents, key-value pairs or a graph. AQL covers filtering, joins, aggregation and graph traversals in one language, and ArangoSearch adds full-text search. The current series is 3.12, at 3.12.13. From 3.12 the source is under the Business Source License 1.1 and Community binaries are under the ArangoDB Community License; from 3.12.5 Community includes all former Enterprise features. The vendor-run option is the Arango Managed Platform (AMP). See our ArangoDB profile.

Side by side

Key differences

Specialist graph database versus multi-model engine

Neo4j is built only for graphs. Relationships are first-class, typed and stored with their own properties, and its tooling (Browser, Bloom, the Graph Data Science library, drivers and connectors) is designed around graph work. If most of your questions are about paths, neighbourhoods and patterns, that focus is an advantage.

ArangoDB treats a graph as documents plus edge collections. The same records can be filtered and aggregated as documents, looked up by key, or traversed as a graph, in one query. That saves running and synchronising a separate document store and graph database, at the cost of a smaller ecosystem and talent pool than single-model leaders such as Neo4j or MongoDB.

Cypher versus AQL

Both languages are declarative but read differently. Cypher describes the shape of the pattern you want; AQL is loop-based, with traversal syntax for graphs. Friends of friends for one person:

// Neo4j Cypher: people two KNOWS hops from Alice
MATCH (:Person {name: 'Alice'})-[:KNOWS*2]->(fof:Person)
RETURN DISTINCT fof.name;

// ArangoDB AQL: the same traversal over the knows edge collection
FOR v IN 2..2 OUTBOUND 'people/alice' knows
  RETURN DISTINCT v.name

In AQL the same query can continue with document-style FILTER, COLLECT and joins to other collections, which is where the multi-model design pays off. Cypher keeps graph patterns concise, especially for variable-length paths and multi-part patterns.

Licences and what the free editions allow

Neo4j Community Edition is GPL v3 and free, but limited in features: it suits single-instance deployments with offline backup by dump, and clustering, online backup, role-based access control and LDAP are Enterprise features. Enterprise Edition is commercial under the Neo4j Software Agreement and priced by sales.

ArangoDB went the other way in 3.12. Its source is under BSL 1.1 (licensor ArangoDB, Inc.), which allows internal production use but not commercial offerings that give third parties database access, and converts each version to Apache 2.0 four years after release. Community binaries are under the ArangoDB Community License, which limits use to internal business purposes with datasets under 100 GB across the cluster; the 3.12 release notes say commercial use or datasets over 100 GiB need a licence, and Arango's pricing page describes Community as free for evaluation and non-commercial use. From 3.12.5 Community includes all Enterprise features. So Neo4j limits the free edition mainly by features, and ArangoDB mainly by use and size. Neo4j's licensing page also describes its developer offerings as generally free for internal, non-production use, so confirm how that applies to your deployment.

Scaling and transactions

Neo4j scales reads and availability through Enterprise clustering or AuraDB. For graphs too large for one machine, Neo4j describes Infinigraph, a specialised Enterprise offering that provides automatic sharding; Community has no sharding. Transactions are ACID and multi-statement.

ArangoDB clusters split collections into shards (hashed on _key by default) with synchronous replication per shard and a Raft-based Agency for cluster configuration; the documentation describes the cluster as CP. SmartGraphs shard graph data so that related vertices and edges stay together. Full ACID transactions apply on a single server and, in a cluster, in OneShard mode, where a database's collections sit on one shard leader. Check whether your transactional needs fit OneShard before planning a sharded cluster.

Managed services, vectors and platforms

AuraDB runs Neo4j on AWS, Google Cloud and Azure with a Free tier and published Professional and Business Critical prices. Neo4j supports vector indexes in both Community and Enterprise, using Lucene's HNSW implementation with up to 4,096 dimensions and cosine or Euclidean similarity, which supports GraphRAG and retrieval next to the graph.

ArangoDB offers the Arango Managed Platform, priced on request. It introduced a vector index type in 3.12.4, enabled with a startup option (named experimental in 3.12.4 and 3.12.5). Since 3.12 it ships Linux packages and Docker images only; production on Windows or macOS hosts is not supported. Neo4j still provides Windows and macOS installs.

Pricing and licensing

Neo4j. Community Edition is free under GPL v3. Enterprise Edition has no published price (contact sales). AuraDB, listed on neo4j.com/pricing on 7 November 2026 with prices shown in dollars (no currency code given): Free at 0, with limits on node and relationship counts; Professional from 0.09 per GB of memory per hour, 1 GB minimum, up to 128 GB memory and 2 TB storage; Business Critical from 0.20 per GB per hour, 2 GB minimum, up to 2 TB memory on AWS and Google Cloud (512 GB on Azure) and 5 TB storage; Virtual Dedicated Cloud by quote. As arithmetic, a 4 GB Professional instance for a 730-hour month is 4 x 0.09 x 730 = about 263 dollars. Neo4j notes that pricing and features are subject to change.

ArangoDB. Community Edition is free only within its licence (internal or non-commercial use, under 100 GB). Arango publishes no prices for Enterprise Edition, the Arango Managed Platform or its platform suites; all are by request.

Pricing checked on the vendors' official pages on 7 November 2026. Prices change; confirm before buying.

Where each one leads

Limitations

When to choose each

When neither is right

• Your data is tabular and aggregate-heavy rather than connected: see SQL vs NoSQL.
• You mainly need a document database with occasional lookups between collections: see MongoDB vs Couchbase.
• You are on Azure and want a managed Gremlin graph API: see our Azure Cosmos DB guide.
• A few recursive queries in an existing relational database may be enough: recursive common table expressions handle simple hierarchies without a graph database.

Final recommendation

Frequently asked questions

Sources

• Neo4j pricing
• Neo4j licensing
• Neo4j database release notes
• Neo4j Operations Manual: introduction and editions
• Neo4j Cypher Manual: vector indexes
• What's new in ArangoDB 3.12
• ArangoDB cluster deployment
• ArangoDB BSL 1.1 licence (GitHub)
• ArangoDB Community License
• Arango pricing

Checked November 2026.

How we research comparisons: our editorial method.