AI-901 Study Guide 2026: Azure AI Fundamentals with Foundry
Everything you need to pass AI-901 — both domains explained, the Azure AI Foundry implementation focus that separates this exam from AI-900, and a 3-week study plan that fits around your day job.
Quick Summary
- • AI-901 is Fundamentals level: 40–60 questions, 60 minutes, 700/1000 to pass, $99 USD
- • Two domains weighted heavily toward implementation (55–60%), not just concepts
- • Replaces and extends AI-900 — adds Azure AI Foundry, Foundry SDK, agents, and Content Understanding
- • Most candidates pass with 2–4 weeks of preparation
- • Requires basic Python literacy (read code, not write large programs)
- • Hardest part: Foundry SDK code-reading questions, system prompt vs. user prompt distinctions, and Content Understanding vs. Document Intelligence scenarios
What Is the AI-901 Exam?
The AI-901 — officially titled Microsoft Azure AI Fundamentals — is Microsoft's implementation-aware fundamentals certification for AI on Azure. It builds on the conceptual foundation of AI-900 by adding a substantial implementation domain centered on Azure AI Foundry: deploying models from the catalog, navigating the Foundry portal, building lightweight chat applications with the Foundry SDK, configuring agents, integrating Azure Speech and Vision, and using Azure Content Understanding for information extraction.
This is not "AI-900 with a different number." More than half of the exam (55–60%) tests Foundry implementation workflows that AI-900 does not cover at all. If you took AI-900 expecting it to satisfy your "I know modern Azure AI" credential goal, AI-901 is the better upgrade.
AI-901 is right for you if you are:
- • A developer, analyst, or product manager working with or alongside Azure AI
- • An IT professional adding AI literacy to your existing Azure skillset
- • A current AI-900 holder updating to the Foundry-centric blueprint
- • A student or career-changer building toward AI-103 or AI-200
- • A non-technical professional in a regulated industry who needs implementation-level credibility on Foundry, not just AI awareness
| Detail | Information |
|---|---|
| Exam Code | AI-901 |
| Full Name | Microsoft Azure AI Fundamentals |
| Questions | 40–60 |
| Time Limit | 60 minutes |
| Passing Score | 700 out of 1000 |
| Price | $99 USD |
| Level | Fundamentals |
| Prerequisites | None formal — basic Python literacy strongly recommended |
| Certification Expiry | Does not expire (Fundamentals) |
Wondering whether AI-901 or AI-900 is right for you? See AI-901 vs AI-900: Which Azure AI Fundamentals Should You Take?
AI-901 Exam Domains & What They Actually Test
AI-901 publishes a two-domain skills outline that hides how front-loaded the exam is on Domain 2. Plan your study time accordingly.
Domain 1: Identify AI Concepts and Capabilities
40–45%- • Responsible AI principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, accountability — and how each maps to concrete Azure features
- • Machine learning fundamentals: supervised vs. unsupervised, classification vs. regression vs. clustering, the difference between training, validation, and test datasets, evaluation metrics (accuracy, precision, recall, F1)
- • Computer vision capabilities: image classification, object detection, OCR, face detection — what each does and which Azure service implements it
- • Natural language processing capabilities: sentiment analysis, key phrase extraction, entity recognition, language detection, translation, summarization
- • Generative AI fundamentals: how transformer-based language models work at a conceptual level, tokens, embeddings, completions, instruction-tuned vs. base models
- • Generative AI risks: hallucination, bias, copyrighted output, prompt injection, model misuse — and the Responsible AI controls Microsoft provides for each
- • Azure AI services overview: Azure OpenAI, Azure AI Foundry, Azure AI Vision, Azure AI Language, Azure AI Speech, Azure AI Search, Azure AI Document Intelligence — knowing which service fits which workload
Domain 2: Implement AI Solutions by Using Microsoft Foundry
55–60%- • Azure AI Foundry portal navigation: hubs vs. projects, model catalog, model card details, deployment workflow
- • Model deployment: deploying an Azure OpenAI model from the catalog (GPT-4o, GPT-4o-mini), open-weight models, Microsoft-published models — and the basic deployment options (region, capacity, content filter)
- • System prompt vs. user prompt vs. assistant message: roles in a chat completion, when each is used, and how system prompts steer behavior
- • Foundry SDK basics: reading Python that creates a Foundry client, deploys a chat completion call, and returns the response — including async patterns
- • Building a chat application: a 10–20 line Python example that sends a user message, gets a model response, and prints the result
- • Foundry agents: creating an agent in the portal, adding tools (file search, code interpreter, function calling), running a thread, and inspecting messages
- • Azure AI Speech in Foundry: speech-to-text and text-to-speech configuration, neural voices, voice selection
- • Azure AI Vision in Foundry: image analysis with captions and tags, basic OCR
- • Azure Content Understanding: defining a custom analyzer schema in natural language, running analysis on documents and images, structured JSON output
- • Choosing between Content Understanding and Document Intelligence prebuilts: Content Understanding for natural-language schema and multimodal sources; Document Intelligence for deterministic structured forms (invoices, receipts, IDs)
- • Responsible AI configuration in Foundry: enabling content filters on a deployment, prompt shields for jailbreak detection, abuse monitoring
- • Authentication and access: API keys vs. Microsoft Entra ID with managed identities, the Cognitive Services OpenAI User role for inference-only access
How Difficult Is AI-901?
Harder than AI-900, easier than AI-103 — and the difficulty distribution is asymmetric. Domain 1 questions look like AI-900 with a slight responsible-AI depth bump. Domain 2 is where most candidates lose points. Three failure patterns dominate:
Skipping the Foundry portal hands-on
AI-901 expects you to recognize Foundry portal screens, the model catalog, deployment options, agent configuration tabs, and Content Understanding analyzer setup. Candidates who only read documentation underperform on questions that show or describe specific portal flows. The fix: spin up a free Foundry hub and project in your own subscription and walk through every workflow once.
Treating Python as optional
AI-901 questions include short Python snippets that use the Foundry SDK. You must be able to read them: what model is being called, what role is being assigned to which message, what the expected return type is. Candidates who skip Python entirely lose 10–15% of the implementation domain to code-reading questions alone.
Content Understanding vs. Document Intelligence confusion
Both extract structured data, but Content Understanding is the right answer when the schema is described in natural language and the source can be multimodal (documents + images + audio + video). Document Intelligence prebuilts (invoice, receipt, ID) are right when the source is a deterministic structured form. Picking the wrong service is a common Domain 2 wrong answer.
Candidates with AI-900 plus basic Python need 2 weeks of focused Domain 2 study. Candidates new to AI and Python need 4–6 weeks with extra time on the Foundry SDK fundamentals.
3-Week AI-901 Study Plan
The Most Tested AI-901 Topics
System Prompts in Azure AI Foundry
A system prompt defines a model's persona, tone, and operating constraints for the entire conversation — applied globally before any user input. Questions test whether you can identify the right place to put behavioral instructions ("respond formally", "stay on-topic", "only answer questions about company policy"). The wrong answer is putting the same instruction in the user prompt or in the deployment settings.
Foundry SDK Code Reading
AI-901 includes Python snippets — typically 10–15 lines — that use the Foundry SDK to call chat completions, configure a project client, or invoke an agent. You must read the snippet and identify what it does, what role is assigned to each message, or which step fails if a specific line is removed. There is no escape from this category by skipping Python.
Content Understanding vs. Document Intelligence
Content Understanding is the natural-language analyzer-based extractor that handles multimodal sources (documents, images, audio, video) with schemas described in plain English. Document Intelligence prebuilts excel at deterministic forms (invoices, receipts, IDs, contracts). Questions test scenario fit — choosing Document Intelligence for a video-content extraction scenario is a common wrong answer.
Responsible AI Principles → Azure Features
Microsoft's six Responsible AI principles map to concrete Azure features. Fairness → Azure Machine Learning fairness toolkit, gender/age-balanced training data. Reliability and safety → content filters, prompt shields, evaluation. Privacy and security → managed identities, private endpoints, encryption. Inclusiveness → accessible UI patterns, multilingual support. Transparency → model cards, attribution. Accountability → audit logs, abuse monitoring. The exam consistently tests principle-to-feature mapping.
Cognitive Services OpenAI Role Selection
Cognitive Services OpenAI User is the least-privilege role for inference. Cognitive Services OpenAI Contributor adds deployment management. Owner adds role-assignment rights. The exam tests this distinction even at Fundamentals level — choose the least-privilege role unless deployment management is explicitly required.
Foundry Agents with Tools
Foundry agents are tested at Fundamentals depth: what an agent is (a persistent assistant with tools, threads, and runs), what tools are available (file search, code interpreter, function calling), and when an agent is the right choice over a single chat completion call. You will not be asked to debug a multi-step agent run, but you will be asked to identify when an agent is the right primitive.
Frequently Asked Questions About AI-901
What is the AI-901 exam?
AI-901 is Microsoft's Azure AI Fundamentals certification focused on implementing AI solutions with Azure AI Foundry. It covers two domains: identifying AI concepts and capabilities (40–45%) and implementing AI solutions using Microsoft Foundry (55–60%). Unlike AI-900, AI-901 expects you to read basic Python that uses the Foundry SDK and to understand the concrete implementation steps for deploying models, building agent applications, and using Content Understanding.
How is AI-901 different from AI-900?
AI-900 tests conceptual understanding of AI and Azure AI services — no coding required. AI-901 goes further: more than half of the exam (55–60%) is implementation-focused. You need to know how to deploy models in Azure AI Foundry, write effective system prompts, build lightweight chat and agent applications using the Foundry SDK, and use Azure Content Understanding to extract structured data from documents and images. AI-901 expects basic Python literacy; AI-900 does not.
How difficult is AI-901?
Harder than AI-900 but still beginner-friendly compared to Associate-level AI-103. The conceptual domain (40–45%) is comparable to AI-900 difficulty. The implementation domain (55–60%) demands familiarity with Azure AI Foundry portal flows and the ability to read Python SDK snippets. Candidates who have AI-900 plus basic Python need 2–3 weeks. Candidates new to both AI and Python need 4–6 weeks.
What does AI-901 cover?
Two domains. Domain 1: Identify AI concepts and capabilities (40–45%) — Responsible AI principles, AI workloads, machine learning fundamentals, computer vision, NLP, generative AI, and a high-level view of Azure AI services. Domain 2: Implement AI solutions by using Microsoft Foundry (55–60%) — Azure AI Foundry portal navigation, model deployment from the catalog, building lightweight chat applications with the Foundry SDK, agent solutions, Azure Speech and Vision integration, and Azure Content Understanding for information extraction.
How long should I study for AI-901?
Most candidates spend 2–4 weeks preparing. If you already hold AI-900 and have basic Python experience, 2 weeks is realistic. Focus on the implementation domain (55–60% of the exam): deploying models in Foundry, building chat and agent applications, using Azure Speech and Content Understanding, and reading Python SDK examples. Candidates new to both AI and coding should plan 4–6 weeks with extra time on Python basics.
Do I need Python for AI-901?
Basic Python literacy is required. You will not write programs from scratch on the exam, but you must be able to read and understand Python that uses the Foundry SDK — variable assignment, function calls, dictionaries, and async/await patterns. If you can read a 10–15 line Python script and explain what it does, you have enough Python. Microsoft Learn has a quick Python primer that covers exactly the syntax AI-901 expects.
What is Azure AI Foundry?
Azure AI Foundry is Microsoft's unified platform for building, evaluating, and deploying AI solutions on Azure. It consolidates the Foundry portal, model catalog (Azure OpenAI, open-weight, and Microsoft-published models), Foundry SDK, agent service, prompt flow, and AI Foundry-integrated Azure AI services (Speech, Vision, Content Understanding, Search). AI-901 tests Foundry-portal workflows and SDK fundamentals — not the advanced platform engineering covered in AI-200.
What is Azure Content Understanding?
Azure Content Understanding is an Azure AI Foundry tool that extracts structured information from documents, forms, images, audio, and video using natural-language analyzer schemas. You describe the fields you want extracted in plain English and the service returns structured JSON. It is tested in the AI-901 information-extraction section. You should know what content types it supports and how it differs from Document Intelligence (which excels at deterministic structured forms).
How much does AI-901 cost?
AI-901 costs $99 USD — the standard Fundamentals exam price. Microsoft regularly issues free or discounted vouchers through Cloud Skills Challenges, Microsoft Learn virtual training days, and partner programs. Check the Microsoft Learn events page before paying full price. AI-901 also appeared in the 2026 Spring Skills Challenge with discount vouchers available to qualifying learners.
Does AI-901 expire?
No. Microsoft Fundamentals certifications do not expire and do not require renewal. Once you pass AI-901, the credential is permanent on your transcript. This makes it a low-risk credential to invest in compared to role-based certifications that require annual free Microsoft Learn renewal assessments.
Is AI-901 worth getting in 2026?
Yes. Azure AI Foundry is Microsoft's primary platform for enterprise AI development and Foundry skills are increasingly listed on job postings for AI engineers, AI product managers, and architects. AI-901 validates that you can navigate Foundry, deploy models, and build basic AI applications — not just talk about AI conceptually. It is a stronger signal than AI-900 alone for any role that involves any hands-on AI work.
How is MSCertQuiz different from free AI-901 practice tests?
Free practice tests for AI-901 typically focus on Domain 1 concepts (which they can recycle from AI-900) and skip the Domain 2 implementation content. MSCertQuiz offers 500 AI-901 questions including Foundry portal scenarios, Foundry SDK Python snippets, agent configuration, system prompt design, and Content Understanding extraction — calibrated harder than the real exam so test day feels easier. Every question includes a full explanation.
Ready to Practice AI-901?
500 practice questions across both domains — Responsible AI, Azure AI Foundry portal and SDK, agents, Content Understanding, and Foundry Python snippets. Start with 40 questions free.
Start Free AI-901 Practice →