AI-901 exam guide: Azure AI Fundamentals after AI-900

AI-901 is the exam for Microsoft Certified: Azure AI Fundamentals since AI-900 retired on 30 June 2026, and it is no longer a concepts-only exam: building with Microsoft Foundry is 55-60% of the syllabus, and Microsoft expects you to know Python. As of October 2026 it lasts 45 minutes, passes at a scaled 700, and costs US$99 in the United States or ₹3,691 in India before tax.

This guide explains the changed syllabus, what to keep from older study material, how much coding to practise, and where to go next. The research was checked on 27 October 2026; the scope follows Microsoft's AI-901 study guide, whose English skills outline is dated 15 April 2026.

AI-901 at a glance

The official exam name is AI-901: Microsoft Azure AI Fundamentals. Passing it earns Microsoft Certified: Azure AI Fundamentals, a fundamentals credential for people starting in AI solution development.

Status: generally available, with no retirement date published. Microsoft's June credentials roundup confirms general availability.
Format, as of October 2026: 45 minutes to answer questions, with 65 minutes of seat time including the surrounding exam activities. Microsoft gives a general range of 40-60 questions across its certification exams; it does not publish an exact AI-901 count or question-type list.
Passing score: 700 on Microsoft's scale of 1 to 1,000. A scaled score is not a percentage of correct answers.
Price, as of October 2026: US$99 in the United States and ₹3,691 in India, before tax. Pricing depends on the country or region where you take the exam.
Prerequisites: no formal prerequisite certification. Python syntax, programming techniques and familiarity with Azure resources are expected.
Delivery: a proctored exam through Pearson VUE, online or at a test centre.
Validity: the fundamentals certification does not expire and needs no annual renewal.

The exam is offered in 13 languages: English, Arabic (Saudi Arabia), Chinese (Simplified and Traditional), French, German, Indonesian (Indonesia), Italian, Japanese, Korean, Portuguese (Brazil), Russian and Spanish. Check the language when you schedule your exam.

You cannot open Microsoft Learn during a fundamentals exam, as of October 2026. Microsoft's exam experience guidance explains the timing and access rules; the sandbox lets you explore the interface beforehand.

What happened to AI-900

AI-900 retired on 30 June 2026, at 11:59 PM Central Standard Time, as recorded in its retired study guide. You can no longer book or retake it. Anyone earning Azure AI Fundamentals now takes AI-901.

If you already earned the certification through AI-900, it stays valid. Microsoft's renewal guidance says fundamentals credentials do not expire, so the exam retirement does not create a renewal deadline for you. There is no upgrade exam you need to pass to keep that credential.

Microsoft's announcement of the change explains that the certification kept its name while the required exam changed. Existing holders need not retake anything; studying the Foundry half is still worth it if you want the new skills.

What changed from AI-900 to AI-901

From five concept domains to two

The retired AI-900 outline divided its scope into AI workloads, machine learning principles, computer vision, natural language processing and generative AI. AI-901 groups the work into identifying AI concepts and capabilities, worth 40-45%, and implementing AI solutions with Microsoft Foundry, worth 55-60%.

The smaller domain count does not mean less to learn. You still need to recognise workloads, then understand how to use those capabilities in an application. Choosing a suitable capability and building a small client for it are different skills, and your preparation needs both.

The new hands-on half: Microsoft Foundry

The current objectives include deploying a model, interacting with it in the Foundry portal, creating a lightweight chat client with the Foundry SDK, and creating and testing a single agent. They also include a client application for that agent and lightweight applications for text, speech, vision and information extraction.

Treat those verbs as your study plan. After each portal exercise, follow the matching client example and explain how the input reaches the model or tool and what the application does with the result. AI-900's outline only asked you to describe or identify these workloads.

The service names to learn

Use the names and scope in the current study guide: Microsoft Foundry, the Foundry SDK, Azure Speech in Foundry Tools, and Azure Content Understanding in Foundry Tools. The outline places model deployment, prompting and agents in Foundry, speech applications under Azure Speech in Foundry Tools, and extraction across documents and other media under Azure Content Understanding in Foundry Tools.

Read each objective as a capability to practise. An older product label in a course does not establish that its exercises cover the current objective, and a similar name is not evidence of a direct service replacement.

Which AI-900 study material still helps

Keep explanations of responsible AI, generative AI, text analysis, speech and computer vision when they help you recognise the workloads in the current outline. For example, distinguishing sentiment analysis from entity extraction remains useful when you start writing a text client.

Use the current objectives to decide what to study next. Older concept-only quizzes leave a gap around model deployment, portal agents and client applications. For each old lesson, ask whether you can connect it to a current objective and demonstrate the relevant task.

The AI-901 syllabus in detail

The syllabus combines choosing an appropriate AI capability with using it in a small solution. The linked exercises below are AI-103 associate-level questions on the same workloads, not AI-901 items; use them to explore the reasoning beyond the fundamentals outline.

Identify AI concepts and capabilities (40-45%)

Start with responsible AI: fairness, reliability and safety, privacy and security, inclusiveness, transparency and accountability. Learn to connect a concern to the design decision it affects. For example, making an application's limitations understandable addresses a different concern from controlling access to sensitive input.

You can extend that thinking with a practice question on content filter severity thresholds. This AI-103 exercise applies safety controls to a deployment. Use it to see how a principle becomes a configuration decision, while keeping the AI-901 objectives as your scope checklist.

Model knowledge includes how generative models work, choosing a model by its capabilities, and understanding deployment options and configuration parameters. Start with the input and output your application needs: generating an image and interpreting an existing image are different requirements. Be able to explain why a model's capabilities fit the task before considering its configuration.

For workloads, distinguish generative and agentic AI, text analysis, speech, vision and information extraction. Text tasks include finding keywords, detecting entities, assessing sentiment and summarising; speech covers recognition and synthesis. Vision includes interpreting images and generating them, while extraction can use text, images, audio or video.

Imagine a support request containing a written description, a photograph and a spoken message. Summarising the description, examining the photograph and recognising the speech require different capabilities even though they support the same customer task. Naming the required output first makes the workload choice easier to explain.

Implement AI solutions by using Microsoft Foundry (55-60%)

For generative AI, practise system and user prompts, model deployment and interaction in the Foundry portal, then follow a lightweight chat client using the Foundry SDK. Create and test a single agent in the portal and work through a client that calls it. Your aim is to explain the purpose of each step and the relationship between the portal configuration and client behaviour.

Prompt construction is easier to understand when you inspect a concrete formatting problem. Try a practice question on few-shot prompt format, an AI-103 exercise about using examples to shape an answer's format. Relate it back to how system and user prompts guide the response in your own Foundry exercise.

For text analysis, build a lightweight application and inspect what its output means. A practice question on key phrases versus named entities helps distinguish a request for topics from a request for named things. It is an AI-103 question, but that distinction supports the text workload selection AI-901 introduces.

Speech preparation includes responding to spoken prompts with a deployed multimodal model and building an application with Azure Speech in Foundry Tools. For a further application choice, try a practice question on fast transcription, which compares approaches to processing finished audio files. Its AI-103 depth is useful extra practice once you understand speech recognition and synthesis.

For vision, practise putting visual input into a multimodal prompt, generating an image, and following a lightweight vision application. A practice question on reading text versus image tagging makes the output distinction concrete: extracting printed words serves a different need from describing what an image contains. Treat this as another AI-103 workload exercise, not a list of service names to memorise for AI-901.

Finally, use Azure Content Understanding in Foundry Tools for information extraction from documents and forms, images, audio and video. Work through a lightweight client application and inspect the extracted information. Reuse the support-request example: decide which information the application must retrieve from each attachment, then check whether the result meets that need.

How much Python you need for AI-901

Microsoft expects knowledge of Python syntax and programming techniques, plus familiarity with REST APIs, SDKs and CLIs. There is no required prerequisite certification, but a course aimed entirely at non-coders will leave you short of the implementation objectives. Start coding alongside the official exercises rather than postponing it until after the concept material.

A useful readiness check is whether you can read a small client, locate its inputs, follow the call to the model or tool, and explain how it handles the response. You should also be able to change a prompt or input and predict what behaviour you are trying to change. These are practical study checks, not a separate Microsoft programming syllabus.

If the sample code is opaque, pause the AI exercise and learn the Python construct you cannot follow, then return to it. If you already write Python, spend more attention on Azure resources, Foundry deployment and the differences between workloads. Existing programming experience helps you work through the examples, but it does not replace learning the AI capabilities.

Cost and how to prepare

As of October 2026, AI-901 costs US$99 in the United States or ₹3,691 in India, before tax, according to Microsoft's exam price feed. Confirm the final price for your exam country when scheduling. These are exam fees; free study material does not make the exam itself free.

If you fail, Microsoft lets you retake the exam 24 hours after the first attempt, with longer waits after that. Each attempt is paid again, and the exam FAQ links the full retake policy.

Begin with the free AI concepts learning path, then work through Get started with AI applications and agents. The AI-901 exam page links to the free practice assessment on AI Skills Navigator and the exam sandbox. Use the assessment to find knowledge gaps and the sandbox to become comfortable with the interface.

Use this suggested three-to-four-week plan as a starting schedule, extending it if Python or Azure is new to you. It is an editorial plan, not a Microsoft estimate of how long passing takes.

First week: concepts and foundations. Complete the concepts path and connect each workload to an example. Practise reading the Python examples and review unfamiliar Azure resource terms.
Next week: Foundry and clients. Deploy and interact with a model, build the chat client, then create and test an agent. Change prompts and explain the results you observe.
Following week: workload applications. Work through text, speech, vision and information extraction exercises. Record which outputs each application produces and which objectives you can now demonstrate.
Final review, with an extra week if needed: take the official practice assessment, revisit weak objectives and try the sandbox. Repeat the practical work you still cannot explain.

There is no AI-901 question bank on getyourcert.org. The AI-103 practice tests support the next, associate-level step, and the AI-103 question list lets you select overlapping workloads for extra practice. Their scores are not an AI-901 readiness measure because the scope and level differ.

For the Azure vocabulary that AI-901 assumes, use the AZ-900 practice tests to review cloud fundamentals. Browse the AZ-900 question list when a cloud concept needs more attention. Pair these supporting resources with the official AI-901 material and your Foundry exercises.

After AI-901: AI-103, AI-200 or AI-300

None of these associate exams requires AI-901. Choose the next subject by the work you want to do; the Microsoft AI certification path explains how the options fit together.

AI-103, Azure AI Apps and Agents Developer Associate: choose it to develop AI applications and agents; the AI-103 exam guide explains the scope.
AI-200, Azure AI Cloud Developer Associate: choose it for the cloud back end around AI, including containers, data, messaging, security and monitoring.
AI-300, Machine Learning Operations Engineer Associate: choose it for MLOps and generative AI operations, including deploying, evaluating and monitoring models. This site has AI-300 practice tests too.

AI-200 is generally available, and its current study guide uses skills measured from 7 October 2026. If your choice extends beyond Microsoft's platform, the best AI certifications guide provides a broader comparison.

Frequently asked questions

Is AI-900 still valid?

The exam retired on 30 June 2026, but an Azure AI Fundamentals certification you earned through it remains valid and does not expire. You do not need an upgrade exam to preserve it. New candidates take AI-901 to earn that same credential.

Is AI-901 hard?

The likely challenge depends on your starting skills: Microsoft expects Python familiarity and Azure resource knowledge, and implementation carries the larger syllabus weight. If you have only studied AI definitions, work through the Foundry exercises before judging your readiness. Use the official practice assessment to identify gaps rather than relying on other candidates' pass stories.

Do I need coding for AI-901?

Yes, you should prepare to understand Python code and lightweight client applications. The objectives include small clients for chat, agents and the other AI workloads. You don't need to be a developer, but you do need to read Python comfortably.

How much does AI-901 cost in India?

As of October 2026, Microsoft's price feed lists ₹3,691 before tax for India. Confirm the final amount when booking, since pricing depends on the country or region where the exam is proctored. The official learning paths and practice assessment are free.

Should I take AZ-900 first?

You do not need AZ-900 as a prerequisite for AI-901. Study its cloud concepts if Azure resource terminology is unfamiliar, then continue with the AI material. Whether to sit that separate exam depends on whether you also want the Azure Fundamentals credential.

Is AI-901 worth it?

It is a useful starting point if you want to build AI solutions on Azure and need a structured introduction to the capabilities and tools. If you already hold Azure AI Fundamentals through the retired AI-900 exam, studying the new implementation material may be more useful than earning the same credential again. If you already build these applications, compare your experience with the associate-level objectives before choosing another fundamentals exam.