AI-103 vs AI-200: Which Microsoft AI Developer Certification Should You Take in 2026?
AI-103 and AI-200 compared domain by domain for 2026: what each Microsoft AI developer exam tests, who it suits, difficulty, cost, renewal, the AI-500 expert path, and which to take first.
Examinotion Team

Last updated: October 2026. Written by the Examinotion team and fact-checked against the official Microsoft Learn certification pages and study guides.
TL;DR AI-103 and AI-200 are both 120-minute Microsoft associate exams with a 700 pass mark and Python as the stated language, but they test different jobs. AI-103 is about building generative AI apps and agents in Microsoft Foundry. AI-200 is about the Azure back end those apps run on: containers, data stores, messaging and monitoring.
If you are a developer choosing between AI-103 (Developing AI Apps and Agents on Azure), which earns the Azure AI Apps and Agents Developer Associate certification, and AI-200 (Developing AI Cloud Solutions on Azure), which earns the Azure AI Cloud Developer Associate certification, you are not alone. Both arrived in 2026, both carry "AI" and "developer" in their names, and both replaced older exams that many people already held. Yet they sit at opposite ends of an AI application. This guide compares the two exams domain by domain, explains who each one suits, and gives a clear recommendation on which to take first.
AI-103 vs AI-200 at a glance
The short answer: AI-103 tests whether you can build the AI part of an application, and AI-200 tests whether you can build and run the cloud services around it. AI-103 centres on Microsoft Foundry, models, retrieval-augmented generation (RAG) and agents. AI-200 centres on containers, Cosmos DB and PostgreSQL vector search, Service Bus, Event Grid, Functions, Key Vault and monitoring [2] [4].
| AI-103 | AI-200 | |
|---|---|---|
| Certification | Azure AI Apps and Agents Developer Associate | Azure AI Cloud Developer Associate |
| Exam title | Developing AI Apps and Agents on Azure | Developing AI Cloud Solutions on Azure |
| Level | Associate (intermediate) | Associate (intermediate) |
| Exam length | 120 minutes | 120 minutes |
| Passing score | 700 out of 1000 | 700 out of 1000 |
| Stated language | Python | Python |
| Main focus | Foundry models, RAG, agents, vision, speech, search | Containers, data services with vector search, messaging, security and monitoring |
| Exam languages | 10 | 13 |
| Free Microsoft Practice Assessment | Yes, on AI Skills Navigator | Not yet available |
| Leads to an expert certification | Yes, Multi-Agent AI Solutions Expert (AI-500) | None listed |
| Renewal | Annual, free online assessment | Annual, free online assessment |
Sources for the table: the AI-103 and AI-200 certification pages and study guides on Microsoft Learn [1] [2] [3] [4], plus the expert certification page [6].
What AI-103 tests: building AI apps and agents in Microsoft Foundry
AI-103 is aimed at an "Azure AI engineer who builds, manages, and deploys agents and AI solutions that take advantage of Microsoft Foundry" [1]. Microsoft expects you to have experience developing apps with Python and to be familiar with generative AI and Azure services [1]. The official skills outline, current as of 16 April 2026, splits the exam into five domains [2]:
| AI-103 domain | Weight |
|---|---|
| Plan and manage an Azure AI solution | 25-30% |
| Implement generative AI and agentic solutions | 30-35% |
| Implement computer vision solutions | 10-15% |
| Implement text analysis solutions | 10-15% |
| Implement information extraction solutions | 10-15% |
The largest domain, generative AI and agentic solutions, covers RAG, agents that use function calling and memory, orchestrated multi-agent solutions, evaluation and tracing [2]. The planning domain is broader than its name suggests: it includes choosing Foundry models and services, CI/CD for Foundry projects, managed identity, private networking, keyless credentials, and configuring safety filters and guardrails [2]. The three smaller domains cover image and video generation, Content Understanding, speech and translation, and semantic, hybrid and vector search with OCR and layout extraction [2].
In practice, AI-103 rewards people who have actually built something in Foundry. If you want a feel for the hands-on side, our Foundry agent tutorial walks through a first agent step by step, and the AI-103 study guide maps each domain to what you need to know.
AI-103 is also the successor to AI-102, the Azure AI Engineer exam that Microsoft retired on 30 June 2026. If you held or were working towards AI-102, our AI-102 retirement guide explains what carries over.
What AI-200 tests: the Azure services behind AI workloads
AI-200 describes a developer who "contributes to all phases" of implementing AI solutions on Azure, with an emphasis on back-end services [3]. Microsoft says candidates should be proficient in Azure SDKs, data management services, messaging and eventing, vector databases, monitoring and troubleshooting, Python, and containerised applications [3]. The skills outline has four domains [4]:
| AI-200 domain | Weight |
|---|---|
| Develop containerized solutions on Azure | 20-25% |
| Develop AI solutions by using Azure data management services | 25-30% |
| Connect to and consume Azure services | 20-25% |
| Secure, monitor, and troubleshoot Azure solutions | 20-25% |
The container domain covers Azure Container Registry and ACR Tasks, App Service containers, Azure Container Apps (including revisions and KEDA scaling) and Azure Kubernetes Service manifests [4]. The data domain is where the "AI" in the title lives: Cosmos DB for NoSQL with vector search and the change feed, Azure Database for PostgreSQL with pgvector and RAG patterns, and Azure Managed Redis for caching and vector indexing [4]. The remaining domains cover Service Bus, Event Grid and Azure Functions, then Key Vault, App Configuration, OpenTelemetry tracing and KQL queries [4].
Notice what is missing. AI-200 does not ask you to build agents, tune prompts, or configure Foundry safety filters. It assumes the model calls exist and asks whether you can store the data, scale the services, connect the events and keep it all secure and observable.
Where the two exams overlap, and where they do not
The overlap is real but narrow. Both exams touch vector search and RAG, but from different directions. On AI-103, vector and hybrid search sit inside the information extraction domain and RAG sits inside the agentic domain, so you are designing how an AI app retrieves grounding data [2]. On AI-200, vector search is a property of the database, so you are configuring Cosmos DB, PostgreSQL with pgvector or Redis to store and query embeddings efficiently [4].
Security and operations also appear on both. AI-103 asks about managed identity, private networking and keyless credentials for Foundry resources [2], while AI-200 asks about Key Vault, App Configuration, OpenTelemetry and KQL across a wider set of Azure services [4].
Everything else is distinct:
- Only on AI-103: Foundry model selection, prompt and agent design, multi-agent orchestration, evaluation and tracing of AI output, safety filters, computer vision, speech, translation, Content Understanding.
- Only on AI-200: container registries, Container Apps, AKS, Cosmos DB request units and change feed, Service Bus, Event Grid, Azure Functions.
That means passing one gives you a head start on perhaps a fifth of the other, not half of it. Treat them as two separate preparation projects.
Which exam fits your role
Microsoft does not publish guidance on which of the two to take first, so the right choice depends on the work you do or want to do.
| If you are... | Start with | Why |
|---|---|---|
| A developer building chatbots, copilots or agents | AI-103 | It tests exactly that work in Foundry |
| A former AI-102 holder or candidate | AI-103 | It is the direct successor |
| A back-end or cloud developer, or a former AZ-204 holder | AI-200 | It tests containers, data and integration work you already do |
| A platform or DevOps engineer supporting AI teams | AI-200 | Hosting, scaling and monitoring are its core |
| Aiming for the Multi-Agent AI Solutions Expert credential | AI-103 | It is the only listed prerequisite [6] |
| A business user or consultant, not a coder | Neither yet | Look at AB-730 or AI-901 first |
On the AZ-204 point: the Azure Developer Associate certification page on Microsoft Learn now states that the certification and its renewal assessment are retired [11]. AI-200 is the new associate-level developer exam covering similar ground (compute, storage, messaging, security and monitoring), now with vector databases added. Our AZ-204 to AI-200 migration guide covers that transition in detail.
If you do not write code at all, both of these exams will be a stretch. Our guide to Microsoft AI certifications without coding and the role-based which certification first guide point to better starting places.
Difficulty and the hands-on skills you need
Both exams are associate level, and neither is a beginner exam. Be realistic: without hands-on time in Azure, either one is hard to pass on knowledge alone.
AI-103 is broader. Five domains span generative AI, agents, vision, speech, translation and search, and the services in this space change quickly. Microsoft renamed Azure AI Foundry to Microsoft Foundry, for example, and the study guide now refers to "Foundry Tools" [2]. Candidates tend to find the breadth, and keeping terminology current, the main challenge.
AI-200 is deeper on infrastructure. Four domains, but each one expects you to know configuration detail: Container Apps revisions and scaling rules, Cosmos DB request units, Service Bus versus Event Grid choices, and KQL queries for troubleshooting [4]. Developers who have run production Azure workloads will recognise most of it. Developers who have only called AI APIs from a notebook will find a lot that is new.
Both exam pages note that the exam may include interactive components [1] [3]. Microsoft's exam experience page also confirms that associate exams allow access to Microsoft Learn during the exam, limited to the learn.microsoft.com domain, with no extra time added and the timer still running [7]. That is a safety net for checking a detail, not a substitute for knowing the material.
Format, cost and renewal compared
The practical details are almost identical:
- Length: 120 minutes for each exam [1] [3].
- Passing score: 700 or greater on Microsoft's scaled score [2] [4]. Our passing score explainer covers what that number actually means.
- Retakes: you can retake after 24 hours following a first failed attempt [1] [3].
- Price: Microsoft states that associate and expert exams typically cost US$165, priced by country or region, with local taxes possibly added [9]. Neither exam page lists a fixed figure, so check the price shown when you book.
- Languages: AI-103 is offered in 10 languages and AI-200 in 13 [1] [3].
- Practice assessment: AI-103 has a free official Practice Assessment, now hosted on Microsoft's AI Skills Navigator and requiring sign-in [1] [10]. AI-200's page says its Practice Assessment is not currently available, and that these usually appear within eight weeks of an exam leaving beta [3].
- Renewal: both are associate certifications that expire every year. Renewal is free, online and unproctored, can be done in a six-month window before expiry, and extends the certification by a year [8].
Should you take both, and in which order?
Taking both makes sense if you build end-to-end AI applications on Azure yourself, or if you lead a small team where one person covers the app and the platform. Together they show that you can build the AI behaviour and run the services it depends on, which is a strong combination for AI engineering roles.
If you plan to take both, start with AI-103 for most developers. There are three reasons:
- It opens the expert path. The Multi-Agent AI Solutions Expert certification lists the Azure AI Apps and Agents Developer Associate as its only prerequisite [6]. AI-200 does not count towards it.
- It has official practice material today. AI-103 already has a free Practice Assessment [10]; AI-200 does not yet [3].
- Its AI concepts make AI-200 easier. Understanding RAG and embeddings from the app side makes the vector database work on AI-200 feel purposeful rather than abstract.
The exception is a cloud or back-end developer with years of Azure experience and little AI work. For you, AI-200 may be the quicker win, followed by AI-103 to add the AI layer.
Where AI-103 leads next: the AI-500 expert path
The Multi-Agent AI Solutions Expert certification is earned by passing AI-500, Designing and Implementing Multi-Agent AI Solutions, and holding the AI-103 associate certification [5] [6]. The AI-500 skills outline covers architecting multi-agent solutions (15-20%), developing them in Azure (30-35%), evaluating, optimising and monitoring them (20-25%), and securing, governing and deploying them (20-25%) [5]. Its audience is expected to be proficient in Python and experienced with Microsoft Foundry, Microsoft Agent Framework, MCP and RAG [5].
AI-200 currently has no expert certification built on top of it. If an expert credential is part of your plan, that settles the order. Our AI-500 explainer covers the expert exam in more depth.
How long to prepare for each exam
Preparation time depends far more on your starting point than on the exam. As a rough guide:
| Starting point | AI-103 | AI-200 |
|---|---|---|
| Already building on Foundry or with Azure OpenAI | 3-4 weeks | 6-8 weeks |
| Experienced Azure back-end developer, new to AI | 6-8 weeks | 3-4 weeks |
| Python developer, new to Azure | 8-10 weeks | 8-10 weeks |
These are Examinotion's planning estimates, not Microsoft figures. Whichever exam you choose, build your plan around the official skills outline and its weightings. For AI-103, our 30-day AI-103 study plan gives a week-by-week schedule, and how to pass AI-103 covers exam-day tactics.
Our recommendation by role
- Building AI apps, copilots or agents: take AI-103. It matches your job and opens the expert path.
- Coming from AI-102: take AI-103. It is the natural successor.
- Coming from AZ-204 or running Azure back ends: take AI-200 first, then consider AI-103.
- Wanting both: AI-103 first, AI-200 second, unless your experience is mostly infrastructure.
- Not sure Azure AI development is for you: start with AI-901 (Azure AI Fundamentals). Our AI-901 vs AI-103 comparison explains when the fundamentals step is worth it.
Frequently asked questions
What is the difference between AI-103 and AI-200?
AI-103 tests building generative AI apps and agents in Microsoft Foundry, including RAG, multi-agent orchestration, vision, speech and search. AI-200 tests the Azure back end for AI workloads: containers, Cosmos DB, PostgreSQL and Redis with vector search, messaging with Service Bus and Event Grid, Azure Functions, and security and monitoring.
Should I take AI-103 or AI-200 first?
Most developers should take AI-103 first. It matches day-to-day AI app work, it already has a free official Practice Assessment, and it is the only prerequisite for the Multi-Agent AI Solutions Expert certification. Experienced Azure back-end developers with little AI background may find AI-200 the quicker first win.
Is AI-200 the replacement for AZ-204?
Microsoft's Azure Developer Associate page states that the AZ-204-based certification and its renewal assessment are retired. AI-200 is the new associate-level developer exam and covers similar ground, including containers, messaging, security and monitoring, with vector databases added. Former AZ-204 holders will find it the closest match to their existing skills.
Is AI-103 harder than AI-200?
Neither is clearly harder; they are hard in different ways. AI-103 is broader, with five domains spanning agents, generative AI, vision, speech and search in fast-changing services. AI-200 is deeper on infrastructure detail such as Container Apps scaling, Cosmos DB request units and KQL. Your existing experience decides which feels harder.
Do I need to know Python for AI-103 and AI-200?
Yes, in practice. Both certification pages name Python as the expected programming language. AI-103 expects experience developing apps with Python, and AI-200 expects proficiency in Python alongside Azure SDKs. You will not write full programs in the exam, but you need to read code and recognise correct SDK usage.
Can I use Microsoft Learn during the AI-103 or AI-200 exam?
Microsoft allows access to Microsoft Learn during associate and expert exams, and both AI-103 and AI-200 are associate exams. Access is limited to the learn.microsoft.com domain, excluding Q&A and Practice Assessments. No extra time is added and the exam timer keeps running, so use it sparingly to check details.
How much do AI-103 and AI-200 cost?
Microsoft states that associate exams typically cost US$165, adjusted by country or region and subject to local taxes. Neither exam page publishes a fixed price, so check the figure shown during booking. Renewal is free: both certifications expire annually and are renewed through an online assessment on Microsoft Learn.
Conclusion
AI-103 and AI-200 look alike on paper, with the same length, passing score and level, but they certify different halves of an AI application. AI-103 proves you can build the AI itself in Microsoft Foundry. AI-200 proves you can build and run the Azure services underneath it. For most developers, AI-103 is the better first step, because it matches modern AI app work, has official practice material and leads to the expert certification.
When you are ready to prepare, start with the official AI-103 exam page and skills outline, then start practising for AI-103 with Examinotion. Our practice tests follow the official domain weightings, with detailed explanations for every answer. If you are weighing other options, browse Examinotion's exam preparation courses to compare every Microsoft AI exam we cover.
Sources
- Microsoft Certified: Azure AI Apps and Agents Developer Associate (AI-103) - Microsoft Learn, accessed 2026-10-07
- Study guide for Exam AI-103: Developing AI Apps and Agents on Azure - Microsoft Learn, accessed 2026-10-07
- Microsoft Certified: Azure AI Cloud Developer Associate (AI-200) - Microsoft Learn, accessed 2026-10-07
- Study guide for Exam AI-200: Developing AI Cloud Solutions on Azure - Microsoft Learn, accessed 2026-10-07
- Exam AI-500: Designing and Implementing Multi-Agent AI Solutions - Microsoft Learn, accessed 2026-10-07
- Microsoft Certified: Multi-Agent AI Solutions Expert - Microsoft Learn, accessed 2026-10-07
- Exam duration and exam experience - Microsoft Learn, accessed 2026-10-07
- Renew your Microsoft Certification - Microsoft Learn, accessed 2026-10-07
- Certification exam FAQ - Microsoft Learn, accessed 2026-10-07
- Practice Assessments for Microsoft Certifications - Microsoft Learn, accessed 2026-10-07
- Microsoft Certified: Azure Developer Associate (AZ-204) - Microsoft Learn, accessed 2026-10-07
Related Articles

AB-620 30-Day Study Plan 2026: Pass the Microsoft AI Agent Builder Associate Exam
A day-by-day 30-day study plan for AB-620, weighted to the official skills outline: lab set-up, agent flows and topics, knowledge and tools, multi-agent and Azure integration, evaluation and ALM, then timed practice.

Microsoft Exam Accommodations and Extra Time: How to Request ESL and Disability Support in 2026
How to request Microsoft certification exam accommodations in 2026: extra time for English as a second language, disability and medical support, the documents you need, the Pearson VUE process, and how online and test-centre delivery differ.

The New AB-730 Agents and Cowork Domain: 2026 Study Guide to the Restructured Exam
Microsoft restructured AB-730 on 20 October 2026 with a new 20-25% agents and Microsoft Copilot Cowork domain. Here is what changed, what the new domain tests, and how to prepare.