- What This Exam Actually Is in 2026
- The Three Domains and Where Your Hours Should Go
- Domain 1: Design and Develop Database Solutions
- Domain 2: Secure, Optimize, and Deploy Database Solutions
- Domain 3: Implement AI and Analytics Capabilities
- Sequencing Your Preparation Around the Domains
- Hands-On Work That Maps to Exam Skills
- Logistics: Scoring, Registration, and Renewal
- Who Benefits Most From This Credential
- Frequently Asked Questions
- DP-420 is now titled Building AI Applications with Azure Cosmos DB, and the certification is Azure Cosmos DB AI Developer Associate.
- The current outline has three domains: design and develop (40-45%), secure/optimize/deploy (30-35%), and AI and analytics (20-25%).
- Partition key design, RU cost reasoning, and change feed behavior underpin questions across all three domains.
- Domain 3 now covers vector, full-text, and hybrid search, agent memory, and Cosmos DB in Fabric.
What This Exam Actually Is in 2026
If you are searching for a DP-420 study plan, start by confirming which credential you are preparing for. Here, DP-420 refers to the Microsoft exam for Microsoft Certified: Azure Cosmos DB AI Developer Associate. Microsoft renamed the certification from Azure Cosmos DB Developer Specialty effective October 6, 2026, and the exam itself is now titled Building AI Applications with Azure Cosmos DB. The exam code did not change. This is the same assigned exam identity and the same existing domain, not a new certificate and not a replacement exam.
That matters for your preparation because older study material, forum posts, and video courses frequently describe a five-domain outline from the earlier Developer Specialty era. The current skills measured are organized into three domains, and the third one, which centers on AI and analytics, is where the rename shows up most visibly. If a resource tells you to memorize five domains with specific point percentages, it is describing the old outline. Use the current structure laid out below, and cross-check any third-party resource against Microsoft's official study guide before you invest time in it.
For a broader orientation, our explainer on what the DP-420 certification is covers the credential's positioning, and the DP-420 requirements guide walks through eligibility and prerequisites.
The Three Domains and Where Your Hours Should Go
The exam is built from three weighted domains. The weights below are the published ranges, and they should drive how you divide your time.
| Domain | Weight | Core Theme |
|---|---|---|
| Domain 1: Design and develop database solutions | 40-45% | Data modeling, partition keys, SDK operations, queries, change feed, AI-assisted development |
| Domain 2: Secure, optimize, and deploy database solutions | 30-35% | Security controls, backup and failover, indexing and performance, monitoring, Fleet |
| Domain 3: Implement AI and Analytics capabilities in database solutions | 20-25% | Vector, full-text, and hybrid search, RAG, agent memory, Cosmos DB in Fabric |
Together, Domains 1 and 2 account for roughly three quarters of the exam. That is the clearest signal about how to prioritize: a candidate who is sharp on modeling, partitioning, RU economics, security, and monitoring but only moderately comfortable with vector search is in a better position than the reverse. That said, Domain 3 is the newest content and the area where candidates with a pure NoSQL background often feel least prepared, so do not skip it. For a deeper breakdown of each content area, see the complete guide to all three DP-420 domains.
Domain 1: Design and Develop Database Solutions
This is the heaviest domain and the one where hands-on SDK experience pays off the most. It spans several clusters of skills.
Resource Models, SDKs, and Account Settings
You need to understand the hierarchy of account, database, container, and item, and how the SDKs surface each one. Questions in this cluster often tie a configuration choice to its consequence.
- Consistency levels and the trade-offs each one makes between latency, availability, and read guarantees
- Request Units as the currency of throughput, and how provisioned versus autoscale versus serverless choices change cost behavior
- Account-level and container-level settings that affect indexing, TTL, and conflict resolution
Queries, Operations, and Concurrency
Expect scenarios that ask which operation is the cheapest or most appropriate for a given access pattern.
- Point reads versus queries, and why a point read by id and partition key is the efficient path
- Paging with continuation tokens and how to resume a query safely
- CRUD, patch operations for partial document updates, and when each is appropriate
- Bulk execution, transactional batch (scoped to a single logical partition), and time-to-live for expiring data
- Optimistic concurrency using ETags
Data Modeling and Partition Keys
This is the intellectual center of the exam. A good partition key distributes storage and throughput evenly and keeps your hottest queries within a single logical partition.
- Embedding versus referencing, and how access patterns decide the answer
- Hierarchical partition keys for workloads that outgrow a single-level key
- Synthetic partition keys when no single property works well
- Multitenant patterns, including how tenant size skew creates hot partitions
- Anti-patterns such as low-cardinality keys, monotonically increasing keys, and designs that force frequent cross-partition fan-out
Change Feed and AI-Assisted Development
- Change feed processing modes, how the change feed processor uses leases to distribute work, and how copy jobs move data
- AI-assisted development workflows, including the Agent Kit, which appears in the current skills outline
A reliable way to internalize this domain is to build one small application end to end and then deliberately break it: pick a poor partition key, watch RU consumption climb, then redesign. For more on how hard the exam feels in practice, read how hard the DP-420 exam is.
Domain 2: Secure, Optimize, and Deploy Database Solutions
The second domain is where operational maturity shows. Many candidates who are comfortable writing code underestimate how much of the exam is about running Cosmos DB safely and efficiently.
Security and Access
Study authentication and authorization as a layered set of choices: keys versus Microsoft Entra ID with role-based access control, and the difference between control-plane and data-plane permissions. Network controls matter as well, including restricting access to approved networks and private connectivity. Dynamic data masking also appears in the skills outline, so know what it protects and what it does not.
Resilience: Backups, Restores, and Failover
- How backup modes work, what a restore produces, and the recovery options available when data is deleted or corrupted
- Multiregion configuration, including multi-region writes and how conflicts are resolved
- Failover behavior, including partition-level failover, and how it differs from a full regional failover
Performance, Indexing, and Observability
Know how the indexing policy affects both query cost and write cost: included and excluded paths, composite indexes for multi-property sorting and filtering, and the trade-off of indexing everything by default. Be able to read a query's RU charge and diagnose why a query is expensive. On the monitoring side, understand metrics and latency, how Azure Monitor and Log Analytics surface diagnostics, and how to investigate throttling.
Fleet
The current outline includes Fleet-level throughput management and analytics across multiple accounts. Understand the purpose: pooling and governing throughput across accounts instead of tuning each in isolation.
Domain 3: Implement AI and Analytics Capabilities
This domain carries the smallest weight but defines the certification's new identity. It is also where fresh study effort tends to deliver the biggest improvement for experienced Cosmos DB developers.
Search: Full-Text, Vector, and Hybrid
Understand when each search type applies and how they combine.
- Full-text search for keyword relevance and ranking over text properties
- Vector search for semantic similarity using embeddings, including vector data types and index types
- Hybrid search that fuses lexical and semantic signals
- Vector size considerations, partitioning, and sharded DiskANN for large-scale vector indexing
- Performance trade-offs, and how all of this supports retrieval-augmented generation (RAG)
Agent Memory
The outline calls out several memory patterns for AI agents backed by Cosmos DB.
- Durable memory that persists knowledge across sessions
- Conversation-state memory that tracks the current interaction
- Semantic memory that retrieves relevant past context by meaning rather than exact match
Be ready to reason about which pattern fits a given agent scenario, and how you would model and partition that data.
Cosmos DB and Cosmos DB in Fabric
- The difference between Cosmos DB and Cosmos DB in Fabric, and when each is the right choice
- Mirroring operational data for analytics without building custom pipelines
- Querying with T-SQL and Spark, and reading and writing data through the Spark connector
Because this content is relatively new, prioritize Microsoft's own documentation and hands-on experimentation over third-party summaries. If you can create a container with a vector index, load embeddings, run a similarity query, and then explain what you would change to scale it, you are well positioned for this domain.
Sequencing Your Preparation Around the Domains
Rather than a generic schedule, order your work by dependency. Partitioning and RU reasoning feed everything else, so they go first. The sequence below assumes a modest weekly commitment and can be stretched or compressed to fit your situation.
Domain 1 foundations
- Work through the resource model, consistency levels, and RU mechanics
- Practice modeling several datasets and justifying a partition key for each
- Run point reads, queries, patches, and transactional batches against a live account
Domain 1 advanced topics
- Implement a change feed processor and observe lease behavior
- Explore hierarchical and synthetic partition keys and multitenant anti-patterns
Domain 2 operations
- Configure Entra ID RBAC, network restrictions, and dynamic masking
- Tune an indexing policy and compare RU charges before and after
- Review backup, restore, and failover options; build Azure Monitor and Log Analytics queries
Domain 3 AI and analytics
- Build a small RAG flow with vector and hybrid search
- Sketch memory designs for an agent and set up a Fabric mirroring scenario
Integration and review
- Take timed practice questions and sort misses by domain
- Revisit weak areas with hands-on repetition, not rereading
When you reach the review stage, a one-page recap helps. Our DP-420 cheat sheet condenses the must-know facts, and the broader DP-420 training overview compares learning formats.
Hands-On Work That Maps to Exam Skills
The exam rewards candidates who have seen Cosmos DB behave, not just read about it. Build a portfolio of small experiments, each tied to a skill the exam measures.
| Experiment | Skill It Reinforces |
|---|---|
| Load data with a deliberately skewed partition key, then redesign it | Partition key selection and hot-partition diagnosis |
| Compare a query against a point read for the same item | RU cost reasoning and operation choice |
| Add a composite index and compare the RU charge of a sorted query | Indexing policy and query performance |
| Process updates through a change feed processor with two consumers | Leases and work distribution |
| Restrict access to a private network and grant a role through Entra ID | Network controls and RBAC |
| Store embeddings and run vector, full-text, and hybrid queries | Search types and RAG patterns |
| Mirror a container into Fabric and query it with T-SQL | Analytics integration |
Key Takeaway
Every experiment above should end with a written sentence explaining why the result occurred in terms of RUs, partitions, or indexes. That habit of explaining the mechanism is what carries over to scenario-based exam questions, where the right answer usually hinges on understanding the underlying cost or distribution behavior.
Logistics: Scoring, Registration, and Renewal
The passing score is 700 on Microsoft's scaled scoring system. A scaled score is not a simple percentage of questions answered correctly, so do not try to reverse-engineer a raw cutoff. Our page on the DP-420 passing score explains how scaled scoring works in more detail.
Be careful with logistics details that vary. The exact current item count, exam duration, and regional fee are not something to take from an outdated blog post. Fees in particular differ by region, so confirm the price for your country on Microsoft's official exam page at the time you register. For the cost picture, see the DP-420 certification cost breakdown, and for scheduling, the DP-420 exam dates guide covers testing windows and deadlines.
Once you are certified, the credential is kept current through an annual renewal. Microsoft provides a free online assessment for this, so you do not retake the full exam each year. Put the renewal window on your calendar so the credential does not lapse.
Who Benefits Most From This Credential
The Associate-level title and the AI focus make this credential a good fit for developers and engineers building data-backed applications on Azure, particularly those adding retrieval, semantic search, or agent capabilities to existing systems. Roles that commonly intersect with the content include application developers, backend and cloud engineers, data engineers working on operational-to-analytical pipelines, and AI application developers who need a reliable persistence and retrieval layer.
Whether the investment makes sense for you depends on your current role and target direction. Our analysis of whether the DP-420 certification is worth it weighs the trade-offs, and the DP-420 jobs overview looks at the kinds of roles where Cosmos DB skills are in demand. If compensation is your focus, the DP-420 salary guide discusses earnings in context.
When you are ready to test yourself under realistic conditions, the DP-420 practice tests let you drill scenario questions by domain and identify where to focus your remaining study time. Working through the full practice question bank after each domain, rather than only at the end, helps you catch gaps while there is still time to fix them.
Frequently Asked Questions
Yes. The exam code remains DP-420. Microsoft renamed the certification from Azure Cosmos DB Developer Specialty to Azure Cosmos DB AI Developer Associate effective October 6, 2026, and the exam is now titled Building AI Applications with Azure Cosmos DB. It is the same assigned exam identity, not a new certificate.
Three: Design and develop database solutions (40-45%), Secure, optimize, and deploy database solutions (30-35%), and Implement AI and Analytics capabilities in database solutions (20-25%). Older five-domain outlines are outdated.
The passing score is 700 on Microsoft's scaled scoring system. Because it is scaled, it does not correspond to a fixed percentage of questions answered correctly.
Start with partition key design and Request Unit reasoning. They appear in Domain 1 modeling questions, Domain 2 performance questions, and even Domain 3 vector partitioning, so early mastery pays off across the whole exam.
Renewal happens annually through Microsoft's free online assessment, so you do not need to retake the full exam. Check your certification dashboard for the renewal window and complete it before the credential expires.
For a complementary read on the numbers behind candidate outcomes, see what the data shows in the DP-420 pass rate analysis, and revisit this DP-420 study guide as the skills outline evolves.