15 Statistics on Graph Query Languages: Cypher, Gremlin, and GQL Adoption

Graph databases are important infrastructure for knowledge graphs, connected enterprise data, and stateful AI systems. Their query-language landscape has historically been fragmented across vendor-specific and community-led approaches. The publication of GQL on April 12, 2024, marked a major change: it was the first new database query language standardized by ISO since SQL in 1987.

For engineering teams, query-language support affects portability, developer experience, and ecosystem fit. It is only one part of the infrastructure decision, however. AI workloads also depend on retrieval quality, temporal context, relationship traversal, metadata controls, and operational architecture.

HydraDB is purpose-built for this broader requirement. It combines graph context, hybrid retrieval, entity resolution, and Git-style temporal versioning in an object-storage-native system for modern AI workloads. Agent memory is one application; teams can also build ontologies, company brains, context graphs, agentic actions, and enterprise knowledge systems while retaining control over graph structure, retrieval, ranking, memory primitives, and context delivery.

Key Takeaways

• GQL is a historic standard: ISO/IEC 39075:2024 is the first new ISO database query language standard in 37 years.

• Cypher has broad implementation support: The openCypher project lists at least 12 vendor, industrial, and research implementations.

• Compatibility is testable: Cypher's Technology Compatibility Kit contains more than 2,000 scenarios designed to validate language behavior across implementations.

• GQL achieved broad consensus: Its final ballot received 14 approvals, no disapprovals, and six abstentions.

• Graph technology spans many use cases: Common categories include fraud detection, recommendations, knowledge graphs, cybersecurity, and dependency analysis.

• AI systems need more than syntax: Query languages express graph operations, while temporal versioning, semantic retrieval, and context assembly depend on the database or platform implementing them.

What Is a Graph Database?

Graph databases represent data as nodes, edges, and properties rather than primarily as rows and tables. This relationship-first model mirrors the way people, systems, events, documents, and concepts connect.

The main elements are:

• Nodes: Entities such as customers, products, services, or documents

• Edges: Connections such as PURCHASED, DEPENDS_ON, or REPORTED_BY

• Properties: Attributes attached to nodes or edges

• Labels or types: Classifications used to organize graph elements

Graph databases are especially useful when an application repeatedly follows relationships across several hops. A query such as “Which engineers worked on services that depend on a deprecated library, and who resolved similar issues?” maps naturally to a graph pattern.

Relational systems can model the same information, and modern relational engines can handle sophisticated recursive queries. Multi-hop access may nevertheless become costly when it requires repeated joins or recursive operations. Actual performance in either model depends on the schema, indexes, query plan, data distribution, traversal depth, and workload.

For AI applications, graphs provide explicit structure around entities and relationships. Temporal knowledge graphs add validity and history, helping an agent distinguish what is true now from what was true earlier.

Why Relationships Matter for AI Workloads

AI applications often need context that is relevant because of a relationship, not merely because its text is similar to a prompt. A support agent may need to connect a customer, an open ticket, the product version involved, and a recent policy change. A coding agent may need to relate a bug to an architectural decision, a dependency, and a prior fix.

Graph-native retrieval makes these connections explicit and traversable. It does not guarantee that every query will be faster than a relational alternative, but it gives developers a direct model for connected reasoning and reduces the need to reconstruct relationships from isolated records at query time.

HydraDB supports relationship-aware retrieval and graph traversal as part of a wider context system. Its retrieval pipeline combines semantic vector search, BM25 keyword search, graph context, and metadata filtering. HydraDB's documentation also reports sub-200ms retrieval latency for its retrieval layer. That figure is a company-reported claim rather than a universal guarantee; observed latency depends on factors such as dataset size, retrieval mode, graph depth, query complexity, and infrastructure.

This unified design is a meaningful advantage for production AI teams. Instead of treating graph traversal, lexical retrieval, semantic retrieval, and persistent context as separate concerns, HydraDB brings them together as composable primitives for context-aware AI.

Cypher: Declarative Pattern Matching for Property Graphs

Cypher is a declarative graph query language closely associated with Neo4j and property graphs. Its pattern syntax visually resembles the graph structure being queried. For example, MATCH (a)-[r]->(b) RETURN a, r, b describes a directed relationship between two nodes.

1. Cypher was first implemented by Neo4j in 2011

Neo4j first implemented Cypher in 2011. Its declarative pattern-matching model established conventions that later influenced other property graph languages and GQL.

2. openCypher has been available since 2015

Neo4j made Cypher available through the openCypher project in 2015. The project provides language specifications, grammar resources, a compatibility test kit, and a process for language improvement.

The open approach allowed other database vendors and research projects to build compatible query interfaces, helping Cypher become a widely recognized property graph language.

3. Cypher's TCK contains more than 2,000 test scenarios

The openCypher Technology Compatibility Kit contains more than 2,000 scenarios. These scenarios define expected query behavior and help implementers test language compatibility.

The TCK should not be read as a general product-reliability benchmark. It validates language behavior, not a database's performance, availability, security, or operational maturity. Even compatible products can differ in supported extensions and execution characteristics.

4. At least 12 systems implement openCypher

The openCypher projects page supports the statement that at least 12 vendor, industrial, and research systems implement or build on openCypher.

This ecosystem gives Cypher practical reach, but compatibility still requires careful evaluation. Teams should test the language features, procedures, extensions, and performance characteristics they actually use.

Gremlin: A Composable Graph Traversal Language

Apache Gremlin is the graph traversal language of Apache TinkerPop. It expresses a traversal as a sequence of composable steps and can run against TinkerPop-enabled graph systems.

Typical step categories include:

• Map steps that transform objects in the traversal stream

• Filter steps that remove objects that do not match conditions

• Side-effect steps that compute or store additional information

• Branch and repeat steps that express conditional or iterative traversal

Gremlin is sometimes described as imperative, while Cypher is described as declarative. That shorthand is useful but incomplete. TinkerPop supports imperative, declarative, and hybrid Gremlin traversals, and execution engines can rewrite traversals through optimization strategies.

The practical distinction is that Cypher centers on graph pattern descriptions, whereas Gremlin centers on a composable traversal pipeline. Cypher can be approachable for common pattern matching, while Gremlin offers granular expression of traversal logic and portability across TinkerPop-enabled systems.

GQL: The ISO Standard for Property Graphs

GQL is an internationally standardized language for creating, accessing, querying, maintaining, and controlling property graphs. It incorporates ideas from earlier languages and research, including Cypher, PGQL, G-CORE, and GSQL.

5. GQL is the first new ISO database query language standard in 37 years

Before GQL, the last new database language standardized by ISO was SQL in 1987. The 37-year interval underscores the importance of establishing a formal standard for property graph data.

The milestone does not mean that existing languages will disappear. Cypher and Gremlin have established ecosystems, and GQL adoption depends on vendor implementation, conformance, tooling, and migration requirements.

6. GQL was published on April 12, 2024

The ISO lifecycle record lists April 12, 2024, as the official publication date for GQL.

Publication gives vendors, buyers, and standards-conscious organizations a formal specification to reference. It does not by itself establish complete compatibility across graph products; conformance still depends on individual implementations.

7. The GQL specification is 610 pages

ISO lists the first edition of the GQL standard at 610 pages. Its length reflects the scope required to define data structures, syntax, semantics, graph operations, types, and edge cases precisely.

8. The GQL standard references more than 400 papers

The GQL specification references more than 400 papers, according to reporting on the standard. This research base reflects the long development history behind graph querying and property graph models.

9. The formal ISO project took about four years and seven months

The ISO lifecycle record shows that the formal GQL project was approved on September 10, 2019, and the standard was published on April 12, 2024. That interval is approximately four years and seven months—not more than five years.

The initiatives behind GQL began earlier. ISO working-group discussions about a standalone graph query language date to at least 2017, before the formal project was approved.

10. The final ballot received 14 approvals and no disapprovals

The GQL Standards chronology reports that the Final Draft International Standard ballot closed with 14 national standards bodies approving, zero disapproving, and six abstaining. This result indicates broad consensus among the voting bodies, although “zero disapprovals” should not be interpreted as universal agreement across every stakeholder in the graph database industry.

11. The DIS review generated 493 comments

The Draft International Standard review generated 493 comments. The GQL Standards chronology says those comments came from three national bodies and the ISO Central Secretariat.

The volume of comments reflects detailed technical and editorial review. It does not, on its own, measure adoption or implementation quality.

12. Seven national bodies nominated experts

Experts were nominated by seven national bodies representing the United States, China, Korea, the Netherlands, the United Kingdom, Denmark, and Sweden. Their participation gave the standard input from multiple regions and technical communities.

Adoption Patterns in Practice

Precise market-share figures for Cypher, Gremlin, and GQL are not publicly established. The available evidence supports a more measured picture:

• Cypher has broad recognition and multiple implementations across graph products and research projects.

• Gremlin remains important in the TinkerPop ecosystem and in systems that value traversal portability.

• GQL provides a formal standard, but production adoption depends on how quickly vendors implement and validate it.

Language choice usually depends on the existing ecosystem, portability needs, query style, team expertise, supported tooling, and standards requirements. For AI workloads, database capabilities such as hybrid search, temporal context, metadata filtering, entity resolution, and context assembly can be more consequential than syntax alone.

HydraDB illustrates this distinction. It lets developers compose graph context, memory primitives, retrieval behavior, ranking controls, and context delivery for stateful AI applications. In HydraDB's company-run benchmark report, using LongMemEval-s and Gemini 3.0 Pro, it scored 97.43% on Knowledge Update, 90.97% on Temporal Reasoning, and 90.79% overall. These are benchmark results, not production guarantees, and the category result should not be presented as overall accuracy.

Open-Source Graph Options

Teams can explore graph technology through several products with different licensing models:

• Apache AGE is an open-source PostgreSQL extension for graph capabilities.

• JanusGraph is an open-source distributed graph database supporting multiple storage backends.

• NebulaGraph is an open-source distributed graph database designed for large graphs.

Query-language support also varies. Some systems implement Cypher or openCypher, some support Gremlin, and GQL conformance is still developing. A proof of concept should verify the exact language surface and operational requirements that matter to the application.

Managed Graph Services

Managed graph services can reduce the operational work associated with provisioning, patching, backups, and availability. Examples include Amazon Neptune, which supports Gremlin and openCypher, and Azure Cosmos DB's Gremlin API.

Teams should assess:

• Query-language and feature support

• Scaling and consistency characteristics

• Data-residency and deployment requirements

• Security controls and certifications

• Integration with existing data and applications

• Pricing and workload predictability

13. One industry guide identifies 10 common enterprise use cases

An industry editorial guide identifies 10 common use cases: fraud detection, recommendations, knowledge graphs, customer 360, master data management, supply chains, IT operations, cybersecurity, access control, and dependency analysis.

This list is a useful taxonomy, not an adoption measurement or proof that every graph database supports every category equally. Product selection still depends on the workload, data model, scale, and operational constraints.

HydraDB offers managed, bring-your-own-cloud, and fully self-hosted deployment paths. Its SOC 2 and ISO 27001 certifications can support vendor due diligence, but certifications do not replace an organization's own security, compliance, data residency, and deployment reviews.

Graph Query Languages and Stateful AI

AI-agent context is an emerging graph database use case. Query languages can express entity patterns and relationship traversals, while production context systems also need to decide what to retain, how to update it, how to rank it, and what to deliver to a model.

Useful capabilities include:

• Entity resolution to connect references with the correct entities

• Relationship tracking to preserve explicit dependencies and associations

• Temporal context to distinguish current facts from historical ones

• Semantic and lexical retrieval to locate relevant unstructured information

• Metadata controls to filter and scope results deterministically

• Provenance to explain where context originated

Temporal versioning and semantic retrieval are platform capabilities rather than properties inherent to Cypher, Gremlin, or GQL. Support varies by implementation. This is why a graph query language alone is not a complete memory or context layer for stateful agents.

14. GQL draws on four established graph languages

The GQL Standards overview explains that GQL incorporates proven ideas from four earlier graph languages: openCypher, PGQL, GSQL, and G-CORE. This synthesis gave the standards group a practical foundation for defining common property graph syntax and semantics.

The influence of those languages does not make their implementations interchangeable with GQL. Each database must still implement and validate the relevant parts of the ISO specification before teams can rely on standards-based portability.

Beyond Vector Similarity

Vector similarity search is valuable for finding semantically related content, but similarity scores do not by themselves encode explicit relationships or temporal validity. Vector databases can store timestamps, metadata, and application-defined connections; using those signals effectively requires suitable schemas, filtering, application logic, or graph-aware infrastructure.

HydraDB combines vector retrieval with lexical, metadata, temporal, and relational graph signals. This gives developers a unified way to retrieve not only content that looks similar, but context connected by identity, history, and relationships. That makes HydraDB a strong fit for persistent agent memory, context graphs, and enterprise knowledge systems where relevance depends on more than embeddings.

15. GQL is designated ISO/IEC 39075:2024

The ISO catalog gives the formal designation as ISO/IEC 39075:2024, Information technology, Database languages, GQL. It provides an authoritative reference for property graph language requirements in procurement, governance, and technical standards.

Formal standardization can improve portability over time, but teams should distinguish between a product's stated direction, partial language support, and verified conformance with the published standard.

Implementation Best Practices

Successful graph deployments begin with the workload rather than the language name.

Data Modeling

• Model important entities and relationships explicitly.

• Use consistent labels, types, and naming conventions.

• Define identity and entity-resolution rules early.

• Plan for temporal validity when facts can change.

Retrieval and Query Design

• Index frequently filtered properties and high-value access paths.

• Bound traversal depth for latency-sensitive queries where appropriate.

• Combine graph context with semantic, lexical, and metadata retrieval when the workload requires it.

• Evaluate relevance and latency using representative data and queries.

AI Context Management

• Separate current state from historical state.

• Preserve provenance for retrieved facts and relationships.

• Scope context by user, team, tenant, or application boundary.

• Measure whether retrieved context improves downstream task performance.

Operations

• Monitor latency distributions rather than averages alone.

• Track graph growth, relationship density, and ingestion lag.

• Test backup, restore, isolation, and access-control behavior.

• Treat vendor benchmarks as inputs to validation, not substitutes for workload-specific testing.

Frequently Asked Questions

What are the main differences between Cypher and Gremlin?

Cypher is a declarative language centered on graph pattern matching. Gremlin is a composable traversal language that can express imperative, declarative, or hybrid traversals. Cypher is often concise for common property-graph patterns, while Gremlin provides detailed traversal composition and portability across TinkerPop-enabled systems. Product-specific extensions and execution behavior still vary.

Why was GQL developed?

GQL provides an international standard for property graph data and operations. It draws on ideas from established graph languages and research while defining a common syntax and semantics. Existing Cypher and Gremlin ecosystems will continue, and GQL's impact will depend on vendor implementation, tooling, and conformance.

How do graph and vector databases compare for AI context?

Vector databases specialize in similarity retrieval over embeddings. Graph databases specialize in explicit entities, relationships, and traversals. Neither category automatically supplies every capability an AI system needs. Timestamps, metadata, ranking, update logic, and context assembly must be designed into the system. HydraDB's advantage is that it combines these retrieval signals with graph context and temporal structure in one platform, addressing common vector database limits.

What are the advantages of a managed graph service?

A managed service can reduce provisioning, patching, backup, and availability work. The trade-offs include vendor dependence, cost structure, deployment constraints, and variations in language support. Certifications help with due diligence but do not remove the need for organization-specific reviews.

How does HydraDB support AI applications?

HydraDB provides graph-native context infrastructure with hybrid semantic and lexical retrieval, graph context, metadata filtering, entity resolution, and temporal capabilities. Developers can use these primitives for persistent memory, ontologies, company brains, context graphs, agentic actions, and other knowledge-intensive systems.