Microsoft AI certification path 2026: which exam first

If you build AI apps, the usual Microsoft AI certification path in 2026 is AI-901 (Azure AI Fundamentals), then AI-103 (Azure AI Apps and Agents Developer Associate), then AI-500 if you want the expert level. If you use AI at work rather than build it, start with AB-730 (AI Business Professional) instead. Microsoft sets no required order for most exams, but those are the two common starting points.

Microsoft replaced most of its AI exams in 2026, and some older roadmaps still list the retired ones. This guide sets out the current Microsoft AI certifications by role: what each exam is, who it is for, how they connect, what it costs, and the order that fits your job. Every exam was checked on Microsoft Learn on October 20-21, 2026.

What changed in Microsoft AI certifications in 2026

Three retirements reshaped the path:

AI-900 became AI-901. Microsoft retired AI-900 on June 30, 2026. AI-901 now leads to the same Azure AI Fundamentals certification, but it is more hands-on: more than half of it is about building with Microsoft Foundry.
AI-102 became AI-103. The last day for AI-102, the exam behind "Azure AI Engineer Associate", was June 30, 2026. Its successor, AI-103, leads to a new certification, Azure AI Apps and Agents Developer Associate, with agents at its centre. The AI-103 exam guide covers that change in detail.
DP-100 gave way to AI-300. The Azure Data Scientist exam retired on June 1, 2026. AI-300, Machine Learning Operations Engineer Associate, is the operations-focused exam that took its place.

Microsoft also added new exams: AI-200 for back-end AI developers, AI-500 for multi-agent systems at expert level, and a business series with "AB" codes for people who use, lead or administer Copilot and agents.

If you searched for "Azure AI Engineer certification", that title belonged to AI-102. You can no longer earn it. AI-103 is the exam to book.

How Microsoft AI certifications are organised

Microsoft uses levels rather than prerequisites for most exams:

Fundamentals and beginner: AI-901 and AB-900 (fundamentals), AB-730 and AB-731 (beginner). All are 45-minute exams.
Associate (intermediate): AI-103, AI-200 and AI-300. Role-based exams for people who build or run AI systems. GH-300, GitHub's Copilot exam, is also rated intermediate.
Expert: AI-500 and AB-100. Both need an associate certification before Microsoft awards the expert credential.

Every exam here passes at 700 on Microsoft's 1,000-point scale. Renewal works differently by level. Fundamentals certifications (AI-901, AB-900) don't expire; Microsoft's pages don't say whether the beginner AB-730 and AB-731 certifications do. Associate and expert certifications expire after one year, and you renew them for free by passing a short online assessment on Microsoft Learn in the six months before expiry. GitHub Copilot is the exception: GitHub certifications are valid for two years for now.

Builder track: AI-901, AI-103, AI-200 and AI-500

AI-901: Microsoft Azure AI Fundamentals

AI-901 is the suggested first step for people who will build AI solutions on Azure, though it is not a prerequisite for anything. It has two domains: identifying AI concepts and capabilities (40-45%) and implementing AI solutions with Microsoft Foundry (55-60%). That second half is the change from AI-900: you are expected to know Python syntax and to be familiar with Azure resources. Fundamentals exams run 45 minutes, and AI-901 is offered in 13 languages. You can't open Microsoft Learn during a fundamentals exam.

AI-103: Azure AI Apps and Agents Developer Associate

AI-103 is the main Microsoft AI certification for developers. It is a 120-minute associate exam in 10 languages, with five domains (skills measured as of April 2026): plan and manage an Azure AI solution (25-30%), implement generative AI and agentic solutions (30-35%), and computer vision, text analysis and information extraction (10-15% each). Microsoft expects Python application development experience.

Agents are the theme. This practice question on choosing a prompt agent over a hosted agent tests the kind of build-or-configure decision AI-103 asks about, and this practice question on Prompt Shields for jailbreak attempts covers the safety side. The AI-103 practice tests give you full sets in the exam's format.

AI-200: Azure AI Cloud Developer Associate

AI-200 (Developing AI Cloud Solutions on Azure) is for back-end developers who build the infrastructure under an AI app rather than the AI features themselves. Its four domains cover containers (Container Registry, Container Apps, AKS), AI data services (Cosmos DB, PostgreSQL with vector search, Managed Redis), connecting Azure services (Service Bus, Event Grid, Functions), and security, monitoring and troubleshooting. Microsoft updated the skills on October 7, 2026 and gave data management more weight: roughly a quarter to a third of the exam. It runs 120 minutes.

AI-500: Multi-Agent AI Solutions Expert

AI-500 (Designing and Implementing Multi-Agent AI Solutions) is the expert step for developers. It 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%), using Microsoft Foundry, the Microsoft Agent Framework, MCP and LangGraph.

Two things to know before planning for it. First, you must already hold the AI-103 certification before Microsoft awards the expert credential. Second, it is new: as of October 2026 it is offered in English only, and its free practice assessment is not available yet.

Operations track: AI-300

AI-300 (Operationalizing Machine Learning and Generative AI Solutions) leads to Machine Learning Operations Engineer Associate. It is the exam for people who deploy, evaluate and monitor models rather than build the app around them. Its five domains are MLOps infrastructure (15-20%), the machine learning model lifecycle (25-30%), GenAIOps infrastructure (20-25%), generative AI quality assurance and observability (10-15%), and optimising generative AI systems (10-15%). It runs 120 minutes and is offered in English only.

It expects Python, entry-level DevOps with GitHub Actions and the CLI, Azure Machine Learning, Foundry, and Bicep or the Azure CLI. This practice question on data drift monitoring is typical of the ML half, and this practice question on an evaluation quality gate is typical of the GenAIOps half. The AI-300 practice tests cover all five domains.

Business track: AB-730, AB-731 and AB-900

The AB exams are for people who use, lead or administer Microsoft 365 Copilot and agents. AB-730 and AB-731 involve no code, and all three run 45 minutes in 13 languages.

AB-730: AI Business Professional

Searches for AB-730 rose sharply in Google Trends over the past 90 days. It is a beginner exam for business users of Microsoft 365 Copilot. Microsoft changed its English skills on October 20, 2026, to four areas: generative AI fundamentals (25-30%), managing prompts and chats in Copilot (20-25%), managing business content and collaboration (20-25%), and driving business outcomes with agents, including Copilot Cowork (20-25%). The agents section is new, so prep material from before late October 2026 will miss it. Microsoft says other languages follow about eight weeks after English.

AB-731: AI Transformation Leader

AB-731 is for business decision-makers who lead AI adoption. It covers the business value of generative AI (35-40%), Microsoft's AI apps and services, including Copilot Studio and Foundry (35-40%), and implementation and adoption strategy (20-25%).

AB-900: Copilot and Agent Administration Fundamentals

AB-900 is for Microsoft 365 administrators. It covers core Microsoft 365 services (30-35%), data protection and governance for Microsoft 365 and Copilot (35-40%), and basic administration of Copilot and agents (25-30%).

Architect track: AB-100

AB-100 leads to Agentic AI Business Solutions Architect Expert, for architects designing AI business solutions across Dynamics 365, Power Platform, Copilot Studio and Foundry. Its three domains are plan (25-30%), design (25-30%) and deploy (40-45%).

AB-100 alone does not earn the credential. You also need one of a list of associate certifications, which includes AI-103 and AI-300 as well as several Power Platform and Dynamics 365 associates. Microsoft updated the English exam on October 14, 2026 and says more updates arrive on November 16, 2026, so check the study guide close to your exam date.

GitHub Copilot: GH-300

GH-300 leads to the GitHub Copilot certification. It is an intermediate, 100-minute exam for developers and DevOps engineers, offered in five languages. It covers using Copilot responsibly (15-20%), Copilot features (25-30%), Copilot's data and architecture, prompt engineering and context, developer productivity, and privacy and content exclusions (10-15% each). It is a GitHub certification, valid for two years, and Microsoft Learn isn't available during GitHub exams.

One exam that rose in the same searches doesn't belong on this list: AZ-802 is Administering Windows Server, not an AI exam.

How the Microsoft AI exams differ

Build vs use: AI-901, AI-103, AI-200, AI-300 and AI-500 are for people who build or run AI systems on Azure. AB-730, AB-731 and AB-900 are for people who use, lead or administer Copilot.
AI features vs plumbing: AI-103 builds the AI part of an app (agents, vision, language). AI-200 builds the cloud services underneath it (containers, databases, messaging).
Build vs operate: AI-103 builds AI apps and agents. AI-300 deploys, evaluates and monitors models and generative AI systems.
Developer expert vs architect expert: AI-500 is hands-on multi-agent development and requires AI-103. AB-100 is solution architecture across Microsoft's business apps and accepts several associate certifications.
Renewal: fundamentals never expire. Associate and expert certifications renew every year with a free online assessment. GitHub Copilot lasts two years.

Which Microsoft AI certification path fits you

Microsoft doesn't require an order for most of these, but these paths build on each other:

AI app developer: AI-901, then AI-103, then AI-500. Skip AI-901 if you already build with Python on Azure.
Back-end or cloud developer: AI-200, then AI-103 if you also want the AI features.
Data scientist or ML engineer: AI-300. Add AI-103 if you build the apps too.
Business user: AB-730. Add AB-731 if you lead an AI rollout.
Microsoft 365 administrator: AB-900, then AB-730.
Solutions architect: AI-103 or a Power Platform or Dynamics 365 associate, then AB-100.
Developer using Copilot daily: GH-300, alongside any of the above.

If you are comparing Microsoft with AWS, Google and others, the best AI certifications guide puts them side by side.

Microsoft AI certification cost and preparation

Microsoft's certification FAQ says associate and expert exams "typically cost US$165", with prices set by country or region as of October 2026. Exam pages show the local price only when you book, so check the price for your country there. Academic discounts exist in most countries, but Microsoft's student discount page excludes India and China. Renewing an associate or expert certification is free.

Microsoft's own preparation is free. Most of these exams have a free practice assessment on Microsoft Learn, which you can retake as often as you like, plus an exam sandbox that shows the question formats. The AI-500 practice assessment is the gap for now.

For practice beyond Microsoft's assessments, this site has full practice tests for the two associate exams most people on the builder and operations tracks take:

The AI-103 question list, including this practice question on RAG chunking strategy.
The AI-300 question list, covering MLOps and GenAIOps.

A workable plan for AI-103 or AI-300: read Microsoft's study guide and mark the skills you haven't used, build each one in a free Azure account, take Microsoft's practice assessment to find gaps, then do timed practice tests until you score well above the pass mark. The RAG vs fine-tuning guide helps with a decision both exams test.

Frequently asked questions

Which Microsoft AI certification should I take first?

If you will build AI apps, AI-901 and then AI-103, or straight to AI-103 if you already code in Python on Azure. If you use AI at work but don't build it, AB-730.

Is AI-901 or AI-103 better?

They are different levels. AI-901 is a 45-minute fundamentals exam that never expires. AI-103 is a 120-minute associate exam for developers that you renew each year. For an Azure AI developer role, AI-103 is the one that matches the job.

Does the Azure AI Engineer Associate certification still exist?

No. It came from AI-102, which retired on June 30, 2026. Its successor is AI-103, which leads to Azure AI Apps and Agents Developer Associate.

How much does a Microsoft AI certification cost?

Microsoft says associate and expert exams typically cost 165 USD, priced by country or region, as of October 2026. The exact local price shows when you schedule the exam. Renewal is free.

Do Microsoft AI certifications expire?

Fundamentals certifications such as Azure AI Fundamentals do not expire. Associate and expert certifications expire after one year, and you renew them free online. GitHub Copilot is valid for two years.

Is a Microsoft AI certification worth it?

If your employer runs on Azure or Microsoft 365, yes: the exams follow the tools you'll use, and the free renewal keeps the certification current. AI-103 is the one with the clearest link to developer roles, and AB-730 is the quickest way to show Copilot skills in a business role.