Microsoft Azure AI Fundamentals (AI‑901) Practice Exams

About the Azure AI-901 exam

Exam at a glance

AI-901 is the exam behind the Microsoft Certified: Azure AI Fundamentals credential. Its objectives are dated 15 April 2026 and they split cleanly in two: name what an AI solution is doing, then build a small one in Microsoft Foundry. Like every Microsoft Fundamentals certification it does not expire, so one pass keeps it on your transcript permanently.

AI-901 replaced AI-900, and it is not a reissue

Microsoft retired AI-900 on 30 June 2026 and AI-901 took its place. The certification name did not change, and anyone already holding it keeps it with no retake. The exam behind it changed a great deal. AI-900 asked you to describe machine learning, computer vision, NLP, and generative AI across five domains. AI-901 collapses the description work into one domain and gives the majority of the exam to hands-on implementation in Microsoft Foundry. If you studied for AI-900, expect the concepts half to feel familiar and the Foundry half to be entirely new ground.

Domain weighting

• Identify AI concepts and capabilities: 40–45% covering responsible AI principles, AI model components and configurations, and the common AI workload families.
• Implement AI solutions by using Microsoft Foundry: 55–60% covering generative AI apps and agents, text and speech, computer vision and image generation, and information extraction.

Who this exam is for

Microsoft frames the candidate as someone at the beginning of a career in AI solution development. That is a shift from the predecessor. AI-901 still welcomes career changers and students, but it now expects you to read Python and to have opened the Azure portal before. It is a strong fit for developers moving into AI, technical support and pre-sales staff who need to build proofs of concept, and analysts who want a credential that proves they can do more than talk about AI.

Prerequisites

None are formally required. Microsoft does assume conceptual knowledge of AI on Azure, Python coding syntax and general programming technique, and familiarity with Azure resources, REST APIs, SDKs, and CLIs. You will not write production code in the exam, but you will be shown short Foundry SDK snippets and asked what they do. AZ-900 is not a prerequisite and most candidates do not take it first.

Why take this certification

• Permanent credential, no renewal. Fundamentals-tier certifications never expire, which beats the one-to-three-year renewal cycle attached to role-based Microsoft, AWS, and Google Cloud certifications.
• It proves you can build, not just describe. The Foundry domain is more than half the exam. Passing AI-901 signals that you have deployed a model, written a system prompt that holds, stood up an agent, and pulled fields off a document.
• It is current. The blueprint is written around Microsoft Foundry, agents, multimodal models, and Azure Content Understanding, so preparing for it teaches you the surface Microsoft is shipping today rather than a legacy service list.
• Clean route to the associate tier. AI-901 leads naturally into AI-103 Azure AI Apps and Agents Developer Associate, which takes the same Foundry building blocks to production depth.

What you'll learn in the AI-901 exam

AI-901 asks two different kinds of question. The concepts domain wants a name for what a scenario describes. The Foundry domain wants the build path, the service, and the parameter. Preparing well means treating them as two separate skills.

Domain 1: identify AI concepts and capabilities (40–45%)

• Responsible AI principles. The six named principles, fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability, and the scenario cues that keep them apart.
• AI model components and configurations. How a generative model produces one token at a time, how the context window is shared between input, output, and reasoning tokens, how to pick a model from its capabilities, and which deployment options and request parameters shape a response.
• AI workloads. Sorting a requirement into generative and agentic AI, text analysis, speech, computer vision, or information extraction, plus the common text analysis techniques such as keyword extraction, entity detection, sentiment analysis, and summarization.

Domain 2: implement AI solutions by using Microsoft Foundry (55–60%)

• Generative AI apps and agents. Writing system and user prompts that do different jobs, deploying a model and exercising it in the Foundry portal, building a lightweight chat client with the Foundry SDK, and creating and testing a single-agent solution.
• Text and speech. Adding text analysis to a small application, responding to spoken prompts with a deployed multimodal model, and building with Azure Speech in Foundry Tools.
• Computer vision and image generation. Interpreting visual input inside a prompt, generating new images with a generative model, and knowing which parameters belong to reading an image versus creating one.
• Information extraction. Using Azure Content Understanding in Foundry Tools to pull fields from documents, forms, images, audio, and video, and understanding the two passes every analysis runs: content extraction, then field extraction against your schema.

The instinct the exam rewards

Two answers often both work. The one that scores is usually the one where durable configuration lives in Foundry and the request carries only what changes this turn. That favors a system prompt over an instruction repeated by every user, one stored agent definition several apps point at over the same instructions pasted into each of them, and a prebuilt analyzer over a schema written from nothing.

Where the boundaries are

Model training, fine-tuning, and MLOps are out of scope and belong to the associate tier. So does deep Azure infrastructure administration beyond what AI work needs. If a question is about producing a model rather than consuming one, it is not an AI-901 question.

How the practice exams help

Every question mirrors the scenario style of the real exam: a short business situation, plausible options, one defensible answer. Each explanation covers why the correct option wins and why each distractor loses, then links the Microsoft documentation behind it. You learn the trade-off, not the answer key.

How to prepare for the AI-901 exam

Budget more time than AI-900 needed. Most candidates with some IT background pass in 3 to 5 weeks, and the extra time nearly all goes into the Foundry domain. A workable sequence:

1. Read the official study guide first (1 day). The AI-901 study guide lists every objective and sub-bullet. Print it and use it as your checklist. Nothing outside it is tested, and everything inside it is fair game.
2. Learn the concepts domain properly (1 week). The six responsible AI principles, the token and context-window model, and the workload families are cheap points that reward memorization. The trick with workloads is to read each scenario for the result that has to come back, not for the material going in. A photograph can lead to a computer vision answer or a generative AI answer depending on what the solution must hand over.
3. Build things in Foundry (2 weeks). This is the part you cannot read your way through. Create a free Azure account, open Microsoft Foundry, and actually do it: deploy a model, run it in the playground, write a system prompt, call the same deployment from a short Python client with the Foundry SDK, stand up a single agent, transcribe a clip with Azure Speech, and run a document through Azure Content Understanding. Half a day per objective is enough, and it converts abstract exam options into things you recognize.
4. Work the free sample, then the premium exams (3 to 5 days). Start with the 10 free questions on this page to calibrate. Then move through the 25 premium exams, reading the explanation on every question you missed and every one you guessed. Use the per-domain breakdown to see whether your gap is in concepts or in Foundry, and aim for a steady 85% on unseen exams before you book.
5. Work through the study guide alongside the questions. The interactive AI-901 study guide is included with premium and is organized by the same objectives, so a weak domain in your results maps straight to the page that fixes it.

Recommended timeline

3 to 4 weeks with some development or cloud background. Allow 6 to 8 weeks if Python and the Azure portal are both new to you, and spend the extra time in Foundry rather than in reading.

Official resources

Start with the AI-901 exam page and its linked study guide. Microsoft's free practice assessment sits on AI Skills Navigator and needs a sign-in. For the Foundry domain, the Microsoft Foundry documentation is the canonical reference for deployments, agents, the SDK, and Foundry Tools.

Frequently asked questions