AI-900 vs AI-901: What Changed in Azure AI Fundamentals (2026)

Short answer: AI-900 retired on June 30, 2026, and AI-901 replaced it. If you are booking Microsoft’s Azure AI Fundamentals exam today, you are taking AI-901. The important part is what kind of change this was. AI-901 is not the same exam with refreshed service names. It is a different exam. AI-900 was five conceptual domains you could pass by knowing which Azure service did which job. AI-901 has two domains, and 55 to 60 percent of it is hands-on implementation in Microsoft Foundry, including reading Python.

So if you have been studying with AI-900 material, a real portion of your prep no longer maps to the exam. That is worth knowing before you book. Here is exactly what changed, taken from Microsoft’s published skills outline dated April 15, 2026.

AI-900 vs AI-901 at a glance

The structural change

AI-900 had five conceptual domains: AI workloads, machine learning principles, computer vision, natural language processing, and generative AI. AI-901 has two:

Read that second row again. The majority of the exam is now implementation. That single line is the difference between a study plan that works and one that does not.

What “hands-on” actually means here

The implementation domain breaks into four areas, all built on Microsoft Foundry (which you may know by its former names, Azure AI Foundry or Azure AI Studio):

Generative AI apps and agents. Writing effective system and user prompts. Deploying a model and working with it in the Foundry portal. Building a lightweight chat client with the Foundry SDK. Creating and testing a single-agent solution. Building a client application for an agent.

Text and speech. Building a lightweight application that includes text analysis. Responding to spoken prompts with a deployed multimodal model. Building an application using Azure Speech in Foundry Tools.

Computer vision and image generation. Interpreting visual input in prompts using a multimodal model. Creating new visual outputs with generative models. Building an application with vision capabilities.

Information extraction. Extracting information from documents and forms using Azure Content Understanding in Foundry Tools. Extracting from images, audio, and video. Building an application with extraction capabilities.

Notice how many of those bullets start with “build”. You are expected to look at code and understand what it does.

What is gone

If your study material spends significant time on these, it was written for AI-900:

• Regression, classification, and clustering as major topics
• Azure Machine Learning designer and automated ML workflows
• Features versus labels, training and validation splits, confusion matrices
• Individual Cognitive Services branding as the main organising idea

These are not banned from appearing as background concepts, but they are no longer where the exam lives.

What stayed

The six responsible AI principles are still there and still worth easy points: fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability. Workload identification survived too, folded into the concepts domain along with text analysis techniques, speech recognition and synthesis, computer vision, and information extraction scenarios. Model configuration is a bigger deal now: know what temperature, top_p, and max tokens actually do.

Three quick checks on your study material

1. Search it for “Foundry.” If Microsoft Foundry is not a major, repeated topic, the material is for the old exam.
2. Search it for “Python.” AI-901 expects you to read code. If there is no code anywhere, that is over half the exam missing.
3. Search it for “Content Understanding.” It is its own sub-domain on AI-901 and was not a focus on AI-900.

You can also go straight to the source. Microsoft publishes the official skills outline with a “skills measured as of” date at the top. For AI-901 that date is April 15, 2026. Anything that predates the change, or that was never rechecked against it, is suspect.

Where our own question bank stands

Our AI-901 bank has been updated and vetted against Microsoft’s current skills outline. It is weighted the way the real exam is weighted, roughly 42 percent concepts and 58 percent Foundry implementation, and it includes questions that put real Python in front of you and ask what the code does, plus full coverage of agents, multimodal models, and Azure Content Understanding. We checked it against all 24 of the individual skill bullets Microsoft publishes, not just the two headline domains.

We also built an automated watcher that checks every certification we cover against its vendor’s published outline and flags us when something changes. Certification vendors refresh exams on their own schedule and rarely announce it loudly, so “when did anyone last check this against the source” is a question worth answering continuously rather than once.

The short version

• AI-900 retired June 30, 2026. AI-901 replaced it.
• It is a replacement, not a refresh. Two domains instead of five.
• 55 to 60 percent is hands-on Microsoft Foundry implementation.
• You need to be able to read Python.
• Foundry, agents, multimodal models, and Content Understanding are the new centre of gravity.
• Regression, clustering, and Azure ML designer are not.

Check the date on whatever you are studying. It matters more this year than it usually does.

Keep reading: How to pass AI-901: a 2-week study plan · Not sure Azure is even your lane? See AWS AI Practitioner vs Azure AI Fundamentals: which should you take first? · or compare all eleven options in Which AI Certification Should You Get?

HOW TO // AI is not affiliated with or endorsed by Microsoft. Azure AI Fundamentals, AI-900, and AI-901 are certifications and trademarks of Microsoft Corporation; we reference them descriptively.