- What the DP-420 Exam Covers in 2026
- The Three Domains at a Glance
- Domain 1: Design and Develop Database Solutions
- Domain 2: Secure, Optimize, and Deploy Database Solutions
- Domain 3: Implement AI and Analytics Capabilities
- Old Five-Domain Outline vs. the Current Three
- Sequencing Your Preparation by Domain
- Where These Skills Show Up on the Job
- Frequently Asked Questions
- DP-420 now leads to Microsoft Certified: Azure Cosmos DB AI Developer Associate, with the exam titled Building AI Applications with Azure Cosmos DB.
- The skills outline has exactly three domains: design and develop, secure/optimize/deploy, and AI and analytics.
- Domain 1 carries the most weight (40-45%), so partition keys, queries, and change feed deserve the most study time.
- Vector search, hybrid search, and agent memory are now testable, not optional extras.
What the DP-420 Exam Covers in 2026
DP-420 is the exam code for Microsoft Certified: Azure Cosmos DB AI Developer Associate. Effective October 6, 2026, Microsoft renamed the credential from Azure Cosmos DB Developer Specialty and retitled the exam Building AI Applications with Azure Cosmos DB. The exam code did not change. This is the same certification identity with a refreshed skills outline, not a brand-new certificate and not a replacement that resets your progress.
If you are new to the credential, start with What Is DP-420 Certification? for background. This guide focuses on the current skills measured: three weighted domains that together describe what a Cosmos DB developer is expected to build, secure, tune, and extend with AI capabilities.
The Three Domains at a Glance
| Domain | Weight | Core Theme |
|---|---|---|
| 1. Design and develop database solutions | 40-45% | Data modeling, partitioning, SDK work, queries, change feed, AI-assisted development |
| 2. Secure, optimize, and deploy database solutions | 30-35% | Security, resilience, performance tuning, monitoring, Fleet |
| 3. Implement AI and analytics capabilities in database solutions | 20-25% | Vector/full-text/hybrid search, agent memory, Fabric integration |
The ranges above are the published weightings for the three areas. Treat them as a guide to proportion rather than a promise about question counts on any given sitting. For the broader picture of how the exam feels, see How Hard Is the DP-420 Exam?
Domain 1: Design and Develop Database Solutions (40-45%)
This is the heaviest domain and the one that most resembles day-to-day application development. It rewards candidates who have actually written code against Cosmos DB, not just read about it.
Resource models, SDKs, and account settings
You need to understand the resource hierarchy (account, database, container, item) and how the SDKs interact with it, plus the settings that shape cost and behavior.
- Consistency levels and the trade-offs each one makes
- Request units (RUs) and provisioned versus other throughput approaches
- Account-level and container-level configuration choices
Working with data: queries and operations
Expect scenarios where you must pick the cheapest, most correct way to read or write data.
- Queries, paging, and continuation tokens
- Point reads versus queries, and why point reads are cheaper
- CRUD operations and partial document updates with patch
- Bulk, batch, and transactional operations
- Time to live (TTL) and optimistic concurrency
Data modeling and partition keys
Partitioning is the concept that separates passing candidates from struggling ones. Know the principles, the patterns, and the failure modes.
- Choosing a partition key for even distribution and efficient queries
- Hierarchical partition keys and synthetic keys
- Multitenant patterns
- Anti-patterns such as hot partitions and cross-partition fan-out
- Embedding versus referencing when modeling relationships
Change feed
The change feed shows up in event-driven and data-movement scenarios.
- Change feed modes and when each applies
- Lease management for distributed processing
- Copy jobs for moving data
AI-assisted development and Agent Kit
New to the current outline, this subtopic reflects how developers now build against Cosmos DB with AI tooling. Review Microsoft's documentation on AI-assisted development workflows and Agent Kit so the terminology is familiar before exam day.
How Domain 1 questions tend to be framed
Microsoft associate-level exams favor scenario-based items: a short business situation, a constraint (cost, latency, consistency, scale), and a decision to make. In this domain that often means choosing a partition key for a described workload, deciding between a query and a point read, or selecting the right change feed approach for a downstream process. Hands-on repetition in the portal and with the SDK pays off more than memorizing definitions.
Domain 2: Secure, Optimize, and Deploy Database Solutions (30-35%)
Domain 2 covers running Cosmos DB responsibly in production. It is less about writing application code and more about governance, resilience, and tuning.
Security
Know the layers of access control and data protection.
- Authentication and authorization approaches for data-plane and management-plane access
- Network controls that limit where traffic can originate
- Dynamic data masking
Availability and recovery
Questions here test whether you can match a recovery or resilience need to the right capability.
- Backups and restores
- Multiregion configuration
- Partition failover behavior
Performance optimization and monitoring
This is where indexing and RU awareness converge.
- Query and index performance, including diagnosing expensive queries
- Metrics and latency analysis
- Azure Monitor and Log Analytics for observability
Fleet
Fleet-level throughput management and analytics across multiple accounts is part of the current outline. Understand what problem it solves for organizations running many Cosmos DB accounts and how it differs from managing each account in isolation.
Reading metrics like an engineer
Expect to interpret scenarios involving elevated latency, throttling, or unexpectedly high RU consumption and then identify the likely cause and remedy. The reasoning chain is usually: confirm the symptom in metrics or logs, locate the offending query or access pattern, then fix it through indexing, query changes, partition design, or throughput settings. If you can narrate that chain out loud, you are ready for this domain's performance questions.
Domain 3: Implement AI and Analytics Capabilities in Database Solutions (20-25%)
The smallest domain by weight is also the one most responsible for the new certification title. Do not skip it because it carries the lowest percentage; it is where candidates with older Cosmos DB experience have the biggest gap.
Search capabilities
Cosmos DB supports several retrieval styles, and you should know when to use each.
- Full-text search, vector search, and hybrid search
- Vector data types, vector indexes, size considerations, and partitioning implications
- Sharded DiskANN for larger vector workloads
- Performance tuning for vector queries
- Retrieval-augmented generation (RAG) patterns that ground model responses in your data
Agent memory
Agentic applications need to remember things. The outline distinguishes several memory roles.
- Durable memory that persists across sessions
- Conversation-state memory for the current exchange
- Semantic memory for retrieving relevant past information by meaning
Analytics and Fabric
Know the difference between running analytics against Azure Cosmos DB directly and using Cosmos DB in Microsoft Fabric.
- Cosmos DB versus Cosmos DB in Fabric
- Mirroring into Fabric
- Querying with T-SQL and Spark
- Reading and writing through the Spark connector
Choosing the right search approach
| Need | Typical Fit |
|---|---|
| Exact or keyword-style matching | Full-text search |
| Meaning-based similarity over embeddings | Vector search |
| Combining keyword relevance with semantic similarity | Hybrid search |
| Grounding an LLM response in stored documents | RAG pattern built on vector or hybrid retrieval |
Scenario questions often describe an application behavior ("results miss obviously relevant items that use different wording") and ask which capability addresses it. Pair each retrieval method with the failure it fixes and you will answer these quickly.
Old Five-Domain Outline vs. the Current Three
If you find older study materials, they may describe a five-domain structure from the Azure Cosmos DB Developer Specialty era. That outline is no longer the reference. The current exam uses the three domains described above, and the content has been reorganized and expanded rather than simply renumbered.
Sequencing Your Preparation by Domain
Because the three domains build on each other, order matters. Partitioning and RU fundamentals from Domain 1 underpin the performance work in Domain 2, and indexing knowledge feeds directly into vector index questions in Domain 3. A sensible progression looks like this:
Domain 1 foundations
- Build a container, experiment with consistency levels and RU costs
- Practice point reads versus queries, patch, bulk, and transactional batch
- Model two or three workloads and defend each partition key choice
- Run a change feed processor and observe lease behavior
Domain 2 operations
- Configure access controls, network restrictions, and masking
- Walk through backup, restore, and multiregion failover options
- Diagnose slow queries using metrics and Log Analytics
Domain 3 AI and analytics
- Create a vector index, load embeddings, and run similarity and hybrid queries
- Prototype a RAG flow and an agent memory store
- Explore mirroring into Fabric and query with T-SQL and Spark
Integration and review
- Take timed practice sets and sort misses by domain
- Revisit weak areas, especially cross-domain scenarios
Adjust the pacing to your background. A developer who already ships Cosmos DB applications can compress Domain 1; someone coming from relational databases should lengthen it. For a fuller plan, see the DP-420 study guide, and keep the DP-420 cheat sheet handy for final review.
Where These Skills Show Up on the Job
The three domains map neatly onto real responsibilities. Domain 1 reflects application and backend developers who model data and write SDK code. Domain 2 reflects the engineers who keep a production Cosmos DB estate secure, observable, and affordable. Domain 3 reflects teams building copilots, search experiences, and agentic features on top of operational data.
Roles that commonly call for this skill set include cloud application developers, backend and data engineers working on Azure, and AI application developers building retrieval-based features. To see how employers describe these needs, browse DP-420 jobs, and for the financial side of the decision, read the DP-420 salary guide and the ROI analysis.
Key Takeaway
Weight your study hours roughly in proportion to the domains, but never leave Domain 3 for last-minute cramming. Vector and hybrid search, agent memory, and Fabric integration require hands-on experimentation, and they are exactly what distinguishes the current exam from its predecessor.
Frequently Asked Questions
Three. They are Design and develop database solutions, Secure, optimize, and deploy database solutions, and Implement AI and analytics capabilities in database solutions. The older five-domain outline no longer applies.
Domain 1, Design and develop database solutions, at 40-45%. Domain 2 follows at 30-35%, and Domain 3 at 20-25%. Spend the most time on data modeling, partition keys, queries, and change feed.
The passing score is 700. For more on how Microsoft scoring works, see DP-420 passing score. Check Microsoft's exam page for current item count, duration, and regional pricing before you register.
Yes. Renewal is annual and done through Microsoft's free online assessment, so you do not need to retake the full exam to keep the credential current.
The rename is the same DP-420 exam identity with an updated title and skills outline, not a separate certificate. Review the current domains closely, since the AI and analytics content is new territory. Details on prerequisites are in DP-420 requirements.