AI-900 Retired Exam Practice & Next Steps — Microsoft Azure AI Fundamentals
Review the final AI-900 outline on iPhone or iPad, preserve the foundational skills you learned, and compare them with the current AI-901 requirements. Core study stays on-device and works offline; optional sync uses your private iCloud account.
The exam
What is the AI-900 exam?
AI-900 was the original exam for Microsoft Certified: Azure AI Fundamentals. Microsoft retired it on 30 June 2026, so it can no longer be scheduled. AI-901 is now the route to the Azure AI Fundamentals credential. This page and its question bank remain available as reference material for the former AI-900 outline.
The exam doesn't ask you to write Python or design a model architecture. It expects a clear conceptual map of AI workload types (computer vision, NLP, document processing, generative AI), the six responsible-AI principles, the difference between regression, classification, and clustering, and which Azure AI service matches a given scenario — Azure AI Vision, Azure AI Language, Azure AI Speech, Azure OpenAI, Microsoft Foundry. The May 2025 outline added significant generative-AI coverage, so a recent question bank matters more than for older fundamentals exams.
Microsoft last updated the AI-900 skills outline on 2 May 2025. Every question in Azure Mastery's AI-900 bank is mapped to that final outline. Read the archived official outline at learn.microsoft.com.
Questions40–60 multiple choice
Duration45 minutes (65 min seat)
Pass score700 / 1000
CostNot schedulable
CredentialRetired exam; earned credentials remain on transcript
StatusRetired 30 June 2026
Skills measured · May 2025
AI-900 exam objectives
Five domains, with weights set by Microsoft's May 2025 update. Every domain summary below is paraphrased from the official skills outline; the bullet-level objectives in Azure Mastery are tagged so you always know which domain you're being tested on.
AI & agent skill mapDataModelsAgentsResponsible AI
AI workloads and considerations15–20%
Explore key topics
The conceptual foundation. Identifies the four common AI workload types (computer vision, natural language processing, document processing, generative AI) with example scenarios for each, and walks through Microsoft's six guiding principles for responsible AI — fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability. Expect scenario questions like "which responsible-AI consideration applies here?" Around 6–12 questions per sitting.
Fundamental principles of machine learning on Azure15–20%
Explore key topics
Common ML techniques (regression, classification, clustering, deep learning, the Transformer architecture) and core concepts (features and labels, training and validation datasets). Covers Azure Machine Learning capabilities: automated ML, the data and compute services, and the model management and deployment story. You don't write code — you identify the right technique for a given scenario. Around 6–12 questions.
Computer vision workloads on Azure15–20%
Explore key topics
Distinguishing image classification from object detection from optical character recognition (OCR) from facial detection — and matching each to the right Azure service. Covers the Azure AI Vision service and the Azure AI Face detection service: what each can do, what each is restricted from doing (Limited Access policies for Face), and which scenario maps to which. Around 6–12 questions.
Natural Language Processing (NLP) workloads on Azure15–20%
Explore key topics
The NLP scenario taxonomy: key phrase extraction, entity recognition, sentiment analysis, language modelling, speech recognition and synthesis, and translation. Maps each scenario to the right Azure tool — Azure AI Language for text analysis, Azure AI Speech for spoken-input/output. Around 6–12 questions.
Generative AI workloads on Azure20–25%
Explore key topics
The largest domain by typical question count, reflecting the May 2025 outline update. Covers what generative AI models are, common scenarios (chat, code, image generation, summarisation), and responsible-AI considerations specific to GenAI. Service coverage includes Microsoft Foundry, Azure OpenAI Service, and the Microsoft Foundry model catalog — what they offer and how they differ. Around 8–15 questions.
Designed for AI-900
How Azure Mastery preserves AI-900 coverage
Azure Mastery ships with 340 AI-900 practice questions, every one written specifically against the current (May 2025) skills outline. Each question carries a domain tag mapped to the official five domains (AI workloads, ML on Azure, computer vision, NLP, generative AI), so you always know which area you're being tested on and where your weak spots are clustered.
The on-device Exam IQ engine can still estimate your performance against the final AI-900 bank for reference. After roughly 30 questions it identifies the specific retired-outline topics that need review, helping you retain foundational knowledge before moving to AI-901.
The adaptive study plan rebuilds itself from your answer history. Get something wrong on object detection vs image classification? You'll see another computer-vision question in the next session. Master "responsible-AI principles" three sessions running and the engine backs off, surfacing fresh generative-AI scenarios. The plan optimises for the gap between where you are and the 700 pass score, not for blind volume.
Knowledge decay tracking is the secret weapon for foundational exams like AI-900. The same domain you mastered six weeks ago is the domain you'll forget by exam day if you stop revising. Azure Mastery tracks every topic's decay curve and flags topics approaching expiry — the padlock icon on the Today screen is your "revisit before you forget" cue.
Historical simulation mode can still run a randomised set from the full 340-question bank with the former 45-minute timing. It is retained for reference and does not represent an exam that can still be booked.
Answer Coach turns each missed answer into a private, grounded lesson: the misconception, key distinction, and rule to remember. It always uses authored certification guidance; on supported devices, an optional on-device model may rewrite the note only when it passes grounding checks.
During your first week, Aura adapts the next step as you go. Every session ends with a concise recap of what changed, what to focus on, and the best follow-up.
Everything essential runs on-device. Your answer history, readiness gauge, and coaching stay private. Optional sync uses your private iCloud account; there is no Azure Mastery account, tracking, or external processing server.
2-week revision plan
Suggested AI-900 study plan
This two-week reference plan reviews the final AI-900 outline before moving to the active AI-901 bank. Use it to retain concepts, not as a booking plan for the retired exam.
Build the mental map
Days 1–2: Tackle AI workloads and considerations. Workload identification + the six responsible-AI principles. 30 questions per session, two sessions per day.
Days 3–4:Fundamental principles of machine learning on Azure. Regression, classification, clustering, deep learning vs Transformer. Light on jargon, heavy on scenario matching.
Days 5–6:Computer vision and NLP together. Service-name discrimination is the trap — Azure AI Vision vs Face vs Document Intelligence; Azure AI Language vs Speech.
Day 7:Generative AI — the largest domain (20–25%). Microsoft Foundry, Azure OpenAI, Foundry model catalog, GenAI-specific responsible-AI considerations.
Sharpen and simulate
Days 8–10: Run the Focus Weak Spots session every morning. The app surfaces the 5–10 questions most likely to move your readiness score.
Days 11–12: Run the Exam Simulator end-to-end at full 45-minute length. Twice. Review carefully after each.
Day 13: Compare the retired outline with AI-901's Microsoft Foundry and Python-example requirements.
AI-900's question bank uses the same formats Microsoft puts on the live exam — not just multiple choice. Each visualisation below is a faithful mock of how the type renders inside Azure Mastery on iPhone and iPad. Exam-simulator mode runs all of them at full 45-minute length with no flag-and-review jumps, mirroring Pearson VUE.
How many principles are included in Microsoft's Responsible AI framework?
4
5
6
8
Multiple choice
A real AI-900 question-bank example with one correct answer. The app explains every option after you answer.
Exam-specific sample
Which models are available through Azure OpenAI Service? (Select all that apply)
GPT-4o
DALL-E
Whisper
Azure SQL
Text-embedding-ada-002
Multi-select
A real AI-900 multi-select item. Every required selection must be correct to earn the mark.
All-or-nothing
Match each AI workload type to its description.
⋮⋮1Classification assigns predefined labels (spam vs.
⋮⋮2Not spam, defect vs.
⋮⋮3Normal) and is supervised.
⋮⋮4Regression predicts continuous numeric values such as price or…
Drag to match
A real AI-900 interactive-format prompt, rendered for touch on iPhone and iPad.
Interactive item
You are reviewing a diagram showing the relationship between AI concepts. Select the concept that is the most specific subset in the hierarchy.
Hotspot
A real AI-900 prompt that tests recognition inside a visual or contextual interface.
Tap target
Azure AI Service Selection Azure AI Language provides a suite of NLP capabilities suited to this pipeline. Language Detection identifies the written language and returns a locale code. Named…
1Which service feature should return the source locale and ISO language code for…
2Which service feature should pull structured fields such as customer name, order…
3Which service feature should score the customer's tone as positive, negative, or…
4Which service feature should learn department labels from example tickets and…
Case studies
A real AI-900 case-study scenario with linked questions that share the same requirements and environment.
Multi-question
✕Your answer: 4
✨ Answer Coach:Four is too few - it leaves out two of the published principles. Microsoft's Responsible AI Standard documents six in total.
— grounded in authored certification guidance
Answer Coach
Answer Coach uses the bank's authored rationale to explain the misconception, key distinction, and rule to remember. On supported devices, an optional on-device model may rewrite the note only when it passes grounding checks.
App exclusive
Frequently asked
AI-900 FAQs
Can I still take the AI-900 exam?
No. Microsoft retired AI-900 on 30 June 2026, so it can no longer be scheduled. AI-901 is now the exam route to the Azure AI Fundamentals credential.
What replaced AI-900?
AI-901 replaced AI-900 on 1 July 2026. It remains the route to Microsoft Certified: Azure AI Fundamentals and adds a more implementation-oriented focus on Microsoft Foundry and Python examples.
What happens if I already passed AI-900?
An Azure AI Fundamentals credential earned through AI-900 remains on your Microsoft Learn transcript. AI-900 itself is retired and cannot be retaken.
Are AI-900 practice questions still useful?
Yes, as reference material for core AI, machine-learning, vision, language, and responsible-AI concepts. New candidates should study the current AI-901 outline and use the AI-901 bank for exam readiness.
AI-900 vs AI-102 — which should I take first?
AI-900 first, unless you already build AI applications professionally. AI-900 is the conceptual map: it teaches you the vocabulary (image classification vs object detection, key phrase extraction vs entity recognition, regression vs classification, the Foundry model catalog) without expecting you to write code. AI-102 is the role-based Associate exam — it expects hands-on Python, REST API calls against Azure AI services, and prompt-engineering against Azure OpenAI. Most candidates pass AI-900 in a few weeks, then spend two to three months on AI-102.
Is AI-900 worth taking if I'm not technical?
Yes — that's a large part of the audience. Microsoft positions AI-900 for both technical and non-technical candidates: solution consultants, sales engineers, technical PMs, business stakeholders evaluating Azure AI procurement, and anyone wanting a credible answer to "which Azure AI service should we use for this?" The exam doesn't ask you to write Python. It asks you to recognise scenarios — "which Azure AI service is best suited for redacting PII from a document?" or "which guiding principle of responsible AI applies here?" — and pick the correct option.
Where AI-900 fits
Certification paths that start with AI-900
AI-900 is the foundational entry point for Microsoft's AI and Data Science role-based tracks. It's optional but strongly recommended — Microsoft markets it as preparation for the Azure AI Engineer Associate and Azure Data Scientist Associate paths, even though it's not a formal prerequisite. If your interest is general Azure infrastructure, look at AZ-900 (Azure Fundamentals) instead.
Azure AI Engineer path
Associate tier
AI-900Fundamentals
AI-102AI Engineer Associate
Azure Data Scientist path
Associate tier
AI-900Fundamentals
DP-100Data Scientist Associate
Generative AI & agent specialisation
Associate / specialty
AI-900Fundamentals
AI-102recommended Associate
orAB-620AI Agent Builder
Generative-AI specialistrole-ready
After AI-900
Related Microsoft certifications
AI-900 is retained as a retired-exam reference. New candidates should use AI-901 for Azure AI Fundamentals, then consider AI-103 for the current role-based Azure AI apps and agents route.
Review AI-900 or move to AI-901
Keep 340 retired-outline AI-900 questions for reference, or use Azure Mastery's current AI-901 bank and adaptive study plan for the active Azure AI Fundamentals route.
