Microsoft · AI-901 Azure AI Fundamentals · Updated November 2026
AI-901 Microsoft Azure AI Fundamentals is the single exam for the Microsoft Certified: Azure AI Fundamentals certification, and it replaced AI-900 when that exam retired on 30 June 2026. It tests two skill areas: AI concepts and capabilities (40–45%) and building AI solutions with Microsoft Foundry (55–60%). You need a scaled score of 700 out of 1,000 to pass, Fundamentals exams run for 45 minutes, Microsoft does not publish the number of questions, and the certification does not expire. This guide covers the format, the skills measured, AI-901 vs AI-900 and AI-103 and a 4-week study plan, and ends with 10 free AI-901 practice questions with answers.
At a glance
| Exam | AI-901: Microsoft Azure AI Fundamentals |
| Certification | Microsoft Certified: Azure AI Fundamentals |
| Passing score | 700 on a 1–1,000 scale |
| Exam time | 45 minutes (65 minutes seat time) |
| Questions | Not published for AI-901 |
| Validity | Does not expire |
- Two skill areas, updated on 15 April 2026: Identify AI concepts and capabilities (40–45%) and Implement AI solutions by using Microsoft Foundry (55–60%).
- Offered in 13 languages, online or at a test centre. Price depends on the country or region where you take it. Free practice: Microsoft’s Practice Assessment on AI Skills Navigator (sign-in required) and the 10 practice questions on this page.
Source: Microsoft Learn AI-901 exam page, the AI-901 study guide and Exam duration and exam experience. Checked on 9 November 2026.
AI-901 Azure AI Fundamentals exam format and scoring
AI-901 is a Fundamentals exam with a 45-minute clock and a passing score of 700 on a scale of 1 to 1,000. Microsoft does not publish the number of questions for AI-901; it says most of its certification exams have between 40 and 60.
Microsoft’s exam duration page gives every Fundamentals exam 45 minutes to answer questions and 65 minutes of seat time. Seat time also covers the instructions, the candidate agreement and any comments. Five minutes of the exam time are set aside for breaks, but the clock keeps running while you are away.
With 40 to 60 questions, 45 minutes gives you roughly 45 to 65 seconds each. You cannot open Microsoft Learn during a Fundamentals exam, unlike associate and expert exams.
| Item | Detail |
|---|---|
| Exam code and name | AI-901: Microsoft Azure AI Fundamentals |
| Certification | Microsoft Certified: Azure AI Fundamentals (AI-901 is the only exam required) |
| Questions | Not published for AI-901; most Microsoft exams have 40–60 |
| Exam time | 45 minutes, including 5 minutes that can be used for breaks |
| Seat time | 65 minutes |
| Passing score | 700 on a scale of 1 to 1,000 (scaled, not a percentage) |
| Question types | Not announced in advance; try them in Microsoft’s exam sandbox |
| Free practice | Microsoft Practice Assessment on AI Skills Navigator (sign-in required); its questions are not the ones on the exam |
| Price | Set by the country or region where you take the exam; no single US$ figure on the exam page |
| Languages | 13, including English, Arabic, Chinese, French, German, Japanese, Korean, Portuguese (Brazil) and Spanish |
| Delivery | Online with a proctor or at a test centre (Pearson VUE; Certiport for students) |
| Validity | Fundamentals certifications do not expire |
| Retirement date | None announced |
Watch out
700 is a scaled score, not 70%. Microsoft’s exam scoring page says it “may not equal 70% of the points”, because the pass mark reflects how hard the questions are. Ignore any site that turns the passing score into a percentage.
There is no penalty for guessing, so answer every question. Some questions may be unscored trial items, and you will not know which ones. For most exams you see your result within minutes, with a score report that shows a bar for each skill area.
Booking, languages and retakes
You book through the “Schedule exam” section of the exam page. Most exams show only Pearson VUE as the exam provider, which offers online proctored exams and test centres; students and academic candidates may see Certiport. If you test online, run the system pre-check on the computer you will use before you register.
Microsoft updated the English version on 15 April 2026 and says localised versions follow about eight weeks after an update. If AI-901 is not available in your preferred language, the study guide says you can request an extra 30 minutes.
Microsoft’s retake policy applies to Fundamentals exams. Microsoft does not refund failed exams, and you pay again for each retake where a fee applies.
| Attempt | Wait before it |
|---|---|
| Second | 24 hours after the first attempt |
| Third to fifth | 14 days between each attempt |
| After five failed attempts | 12 months from your first attempt |
AI-901 skills measured: the two domains and their weights
The AI-901 study guide, with skills measured as of 15 April 2026, has just two domains. Building with Microsoft Foundry carries 55–60% of the exam, so more than half your marks come from hands-on tasks, not definitions.
- 1. Identify AI concepts and capabilities (40–45%): Responsible AI principles; how generative AI models work, choosing a model and its configuration parameters; recognising AI workloads such as generative and agentic AI, text analysis, speech, computer vision and information extraction.
- 2. Implement AI solutions by using Microsoft Foundry (55–60%): Prompts, model deployment and a chat client with the Foundry SDK; single-agent solutions; text and speech apps with Azure Speech in Foundry Tools; vision and image generation with multimodal models; information extraction with Azure Content Understanding in Foundry Tools.
Identify AI concepts and capabilities (40–45%)
Responsible AI. You need Microsoft’s six principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability.
Model components and configuration. You should be able to explain how generative AI models produce output, pick a model by its capabilities (for example text only or text plus images), and know what deployment options and parameters such as temperature or maximum tokens do.
AI workloads. This covers generative and agentic AI, text analysis (keyword extraction, entity detection, sentiment analysis and summarisation), speech recognition and synthesis, computer vision and image generation, and extracting information from text, images, audio and video.
Implement AI solutions by using Microsoft Foundry (55–60%)
This domain is about doing the work in Microsoft Foundry, with small amounts of Python. Microsoft says you should know Python syntax and be familiar with REST APIs, SDKs and CLIs.
Generative AI apps and agents. Write effective system and user prompts, deploy a model and test it in the Foundry portal, build a lightweight chat client with the Foundry SDK, and create and test a single-agent solution with a simple client app.
Text and speech. Build a small app that includes text analysis, respond to spoken prompts with a deployed multimodal model, and use Azure Speech in Foundry Tools.
Vision and image generation. Send images in prompts to a multimodal model, create new images with generative models, and build a small app with vision features.
Information extraction. Use Azure Content Understanding in Foundry Tools to pull information from documents and forms, images, audio and video, and build a small app around it.
Exametics practice tests
AI-901 Practice Tests: 6 tests, 360 questions
- 6 practice tests of 60 questions each, including a timed Exam Mode test
- Every answer explained: why the right answer is right and why the others are wrong, with a hint for each question
- Each question mapped to an official AI-901 skill area
- Current product names such as Microsoft Foundry and Content Understanding
- Lifetime access
30-day refund. Instant access after payment.
Who should take AI-901 and what you need first
AI-901 is for people at the start of a career in AI solution development. There is no prerequisite exam or certification, but Microsoft expects Python basics and some familiarity with Azure resources.
Microsoft’s audience profile asks for “conceptual knowledge of AI solutions in Azure” and the foundational technical skills to work with them. That is a change from AI-900, which Microsoft described as suitable for technical and non-technical backgrounds with no software engineering experience required.
It suits students, junior developers and career changers who can read simple Python. If you have never written code, plan extra time to follow a short SDK example.
Microsoft offers an instructor-led course, Introduction to AI in Azure (AI-901T00). Its audience note recommends core cloud concepts, including storage, compute and cloud authentication. If those are new to you, our AZ-900 Azure Fundamentals guide covers the cloud basics.
AI-901 vs AI-900 and AI-103: which first?
If you are starting with Azure AI, take AI-901: AI-900 retired on 30 June 2026 and you can no longer book it. AI-103 is the intermediate, developer-level exam to take next. AI-901 is not required before AI-103.
| Exam | Status and level | Key facts |
|---|---|---|
| AI-900: Microsoft Azure AI Fundamentals | Retired on 30 June 2026; replaced by AI-901 | Five areas (AI workloads, machine learning, computer vision, NLP, generative AI); passing score 700 |
| AI-901: Microsoft Azure AI Fundamentals | Current Fundamentals exam; no prerequisite | Two areas (AI concepts 40–45%, Microsoft Foundry 55–60%); 45 minutes; passing score 700; does not expire |
| AI-103: Developing AI Apps and Agents on Azure | Intermediate; leads to Azure AI Apps and Agents Developer Associate | Five areas, led by generative AI and agentic solutions (30–35%); 120 minutes; passing score 700; renew yearly |
Sources: AI-900 exam page, Azure AI Apps and Agents Developer Associate and the AI-103 study guide. Checked on 9 November 2026.
The biggest change from AI-900 is the hands-on focus. AI-900 asked you to describe workloads and had a separate machine learning area. AI-901 drops that area and puts more than half the exam on building with Microsoft Foundry. If you passed AI-900, Microsoft says retirement does not invalidate a certification you have already earned.
AI-103 goes much deeper: planning and securing Foundry solutions, retrieval-augmented generation, multi-agent orchestration and monitoring. Microsoft lists Python app development experience for it. Doing AI-901 first gives you the vocabulary and the Foundry basics that AI-103 assumes. Our AI-103 Azure AI Engineer practice tests guide explains what that exam adds.
Next on your Microsoft path
- AI-103 Developing AI Apps and Agents on Azure practice tests: the developer-level next step after AI-901, built on the same Foundry tools.
- AZ-900 Microsoft Azure Fundamentals practice tests: cloud basics that help with the Azure side of AI-901.
- AB-900 Microsoft 365 Copilot and Agent Administration Fundamentals practice tests: a second Fundamentals exam if your agents will live in Microsoft 365 (see our AB-900 exam tips).
A 4-week AI-901 study plan
Four weeks of steady study is a realistic plan if you know basic Python and have an Azure subscription to practise in. Give most of your time to the Foundry domain, because it carries more than half the marks. Before you start, try the 10 free practice questions further down this page to see where you stand.
- Week 1: AI concepts and responsible AI Learn the six responsible AI principles with an example of each. Study how generative models work, what tokens are, and what temperature and maximum tokens change. Map each workload (text analysis, speech, vision, information extraction, agents) to a real use.
- Week 2: Models, prompts and agents in Foundry Deploy a chat model in the Foundry portal and test it in the playground. Write system and user prompts and see how they change answers. Build a short Python chat client with the Foundry SDK, then create and test a single agent with a knowledge or tool connection.
- Week 3: Text, speech, vision and Content Understanding Build small apps: text analysis on customer reviews, speech to text and text to speech with Azure Speech in Foundry Tools, image questions with a multimodal model, and an image generated from a prompt. Then extract fields from a sample invoice, an image and an audio file with Content Understanding.
- Week 4: Timed practice and review Take the free Microsoft Practice Assessment, then full AI-901 practice tests in timed Exam Mode, aiming to finish inside 45 minutes. Read every explanation, including for questions you got right, and revisit the skill area where you score lowest.
Tip
Try the Microsoft exam sandbox before exam day to see the question types and the review screen. Remember that once you take a break, you cannot go back to questions you have already seen, so answer or review them first.
Is AI-901 hard?
AI-901 is a Fundamentals exam, but it is more practical than AI-900. More than half of it is about building with Microsoft Foundry, and Microsoft expects Python basics, so people who only read about AI concepts will find it harder. With some hands-on time in Foundry, about four weeks is a realistic plan for a beginner who can read simple Python.
Planning more than one exam?
If you plan to follow AI-901 with AI-103, AZ-900 or another Microsoft exam, note that Exametics has 21 Microsoft practice-test sets. The All-Access Pass covers every practice test on Exametics, including new exams as they are added, with lifetime access. It costs less than buying about six single practice tests.
Free AI-901 practice questions (10 with answers)
Try these 10 free AI-901 practice questions, then open each answer to check your reasoning. They are original Azure AI Fundamentals sample questions, spread across the two official domains by weight; allow about 10 minutes. The full Exametics AI-901 practice tests have 360 more in the same style.
Practice question 1 · Identify AI concepts and capabilities
A team adds live captions and screen-reader support so that people with hearing or sight impairments can use its AI assistant. Which responsible AI principle does this support most directly?
- A. Fairness
- B. Transparency
- C. Inclusiveness
- D. Accountability
Show answer
Answer: C. Inclusiveness means designing AI so it empowers and engages everyone, including people with disabilities. Captions and screen-reader support make the assistant usable by more people.
A is wrong because fairness is about the system treating people and groups equally, for example not approving loans at different rates for similar applicants. B is wrong because transparency is about people understanding how the system works and its limits. D is wrong because accountability is about people being answerable for how the AI system operates.
Practice question 2 · Implement AI solutions by using Microsoft Foundry
Every conversation in your chat app must follow the same rules: answer only questions about your company’s products, in a friendly tone. Where should you put these instructions?
- A. In the system message
- B. In each user message
- C. In an assistant message
- D. In a stop sequence
Show answer
Answer: A. The system message sets the model’s role, rules and tone for the whole conversation, so it is the right place for instructions that apply every time.
B is wrong because user messages carry what the user asks; repeating rules there is unreliable and users can change them. C is wrong because assistant messages hold the model’s earlier replies, not its instructions. D is wrong because a stop sequence only tells the model where to stop generating text.
Practice question 3 · Identify AI concepts and capabilities
A support assistant gives noticeably different wording each time it answers the same question. You want more consistent, focused answers. Which configuration change should you make?
- A. Raise the temperature
- B. Raise the maximum number of tokens
- C. Add more stop sequences
- D. Lower the temperature
Show answer
Answer: D. Temperature controls how random the model’s token choices are. A lower temperature makes the model favour the most likely tokens, so answers are more consistent and focused.
A is wrong because a higher temperature makes output more varied and creative, the opposite of what you want. B is wrong because maximum tokens only limits how long a response can be. C is wrong because stop sequences only end the output early when a set text appears.
Practice question 4 · Implement AI solutions by using Microsoft Foundry
You need to pull the vendor name, invoice date and total from thousands of scanned invoices and return them as structured fields. Which Foundry Tools service fits best?
- A. Azure Speech in Foundry Tools
- B. Azure Content Understanding in Foundry Tools
- C. Azure Translator in Foundry Tools
- D. An image-generation model
Show answer
Answer: B. Content Understanding extracts information from documents and forms, as well as images, audio and video, and can return the fields you define as structured output.
A is wrong because Azure Speech works with spoken audio, not scanned documents. C is wrong because Translator changes text from one language to another; it does not extract fields. D is wrong because image-generation models create new images rather than reading existing ones.
Practice question 5 · Identify AI concepts and capabilities
An app reads news articles and labels each person, organisation and location mentioned in them. Which text analysis technique is this?
- A. Sentiment analysis
- B. Entity detection
- C. Keyword extraction
- D. Summarisation
Show answer
Answer: B. Entity detection, also called named entity recognition, finds items in text and classifies them into types such as person, organisation and location.
A is wrong because sentiment analysis judges whether text is positive, negative, neutral or mixed. C is wrong because keyword extraction returns the main talking points of a text without classifying them by type. D is wrong because summarisation produces a shorter version of the text.
Practice question 6 · Implement AI solutions by using Microsoft Foundry
You are building a single agent in the Foundry portal. It must answer staff questions using the content of several HR policy PDF files that you upload. Which tool should you add to the agent?
- A. Code interpreter
- B. Function calling
- C. File search
- D. Image generation
Show answer
Answer: C. The file search tool indexes the files you upload and lets the agent retrieve relevant passages to ground its answers in your own documents.
A is wrong because code interpreter runs Python code, for example to analyse data or draw charts; it is not meant for searching document content. B is wrong because function calling lets the agent call functions you define and host, which you would have to write yourself. D is wrong because image generation creates pictures and cannot answer questions from documents.
Practice question 7 · Identify AI concepts and capabilities
How does a large language model produce its reply to a prompt?
- A. It repeatedly predicts the next token, based on the prompt and the tokens it has generated so far
- B. It looks up a stored answer to the same question in a database
- C. It fills in a fixed template written by a developer for each topic
- D. It searches the internet and copies the best matching page
Show answer
Answer: A. A language model generates text one token at a time. At each step it uses the patterns learned in training to predict a likely next token, given the prompt and everything generated so far.
B is wrong because the model does not store and retrieve fixed answers; that is why the same prompt can give different wording. C is wrong because templates describe rule-based bots, not generative models. D is wrong because a model does not browse the web by itself; it can only use web results if an app or tool supplies them.
Practice question 8 · Implement AI solutions by using Microsoft Foundry
Your Python chat client calls a deployed model through the chat completions API. Users complain that the model forgets what they said two messages ago. What should the client do?
- A. Raise the temperature
- B. Raise the maximum number of tokens in the response
- C. Deploy a larger model
- D. Send the earlier user and assistant messages with each new request
Show answer
Answer: D. The chat completions API does not remember earlier requests. The client keeps the conversation history and sends it in the messages list with every call, so the model can see the earlier turns.
A is wrong because temperature changes how random the output is, not what the model can see. B is wrong because maximum tokens limits the length of the reply, not the context sent in. C is wrong because a larger model still only sees the messages included in the request.
Practice question 9 · Implement AI solutions by using Microsoft Foundry
You are building a delivery app with Azure Speech in Foundry Tools. The app must read order updates aloud to customers in a natural-sounding voice. Which feature should the app call?
- A. Speech to text
- B. Text to speech
- C. Speech translation
- D. Sentiment analysis
Show answer
Answer: B. Text to speech, also called speech synthesis, turns written text into spoken audio, using neural voices that sound natural.
A is wrong because speech to text (speech recognition) does the reverse, turning spoken audio into text. C is wrong because speech translation converts speech from one language into another. D is wrong because sentiment analysis is a text analysis feature that detects opinion; it does not produce audio.
Practice question 10 · Implement AI solutions by using Microsoft Foundry
In your app, a user uploads a photo of a damaged parcel and types “What is wrong with this parcel?” What is the simplest way to answer?
- A. Send the text to a text embedding model
- B. Send the text to an image-generation model
- C. Send the image and the question together to a deployed multimodal model that accepts image input
- D. Send the image to speech to text
Show answer
Answer: C. A multimodal model can take an image and text in the same prompt and answer questions about what it sees, which is exactly this task.
A is wrong because an embedding model turns text into vectors for search or similarity; it does not look at images or write answers. B is wrong because an image-generation model creates new images instead of describing an existing one. D is wrong because speech to text works on audio, not photos.
How did you score?
8 to 10 correct: you are ready for a full timed practice test inside 45 minutes. Under 8: each question shows its official domain, so revise the domain you missed most, then try again. The AI-901 practice tests have 360 more questions like these, each with a full explanation.
AI-901 dumps vs practice tests
AI-901 dumps are question sets that claim to be copied from the live exam, and using them can cost you your certification. Practice tests that explain every answer are the safer way to prepare, because they build the reasoning the exam tests.
Before the exam, you accept the Microsoft Certification Exam Candidate Agreement, which requires you to keep exam content confidential. Microsoft’s exam security policy says that using brain dump material to prepare counts as cheating. Microsoft can revoke your scores and certifications and ban you from future exams.
Dumps are also unreliable. Their answers are often wrong or out of date, and AI-901 changed a lot from AI-900, with new product names such as Microsoft Foundry and Content Understanding.
Good practice tests use original questions mapped to the official skills, explain every option and let you rehearse under a timer. Even Microsoft’s free Practice Assessment says its questions are not the same as those on the exam; the 10 free practice questions above follow the same approach.
Microsoft Azure AI Fundamentals (AI-901) FAQs
How many questions are on the AI-901 exam?
Microsoft does not publish the number of questions for the AI-901 Microsoft Azure AI Fundamentals exam. Its exam duration page says most Microsoft Certification exams have between 40 and 60 questions and that Fundamentals exams last 45 minutes.
What is the passing score for AI-901?
The AI-901 exam page gives a passing score of 700 on a scale of 1 to 1,000. Microsoft’s scoring page explains that this is a scaled score, so it may not equal 70% of the points.
Are there free AI-901 practice questions?
Yes. This guide includes 10 free AI-901 practice questions with answers and explanations, each labelled with its official domain, and Microsoft offers a free Practice Assessment on AI Skills Navigator (you must sign in to launch it). For full timed practice, the Exametics AI-901 practice tests have six 60-question tests (360 questions) with every answer explained.
Is AI-901 hard, and how long should I prepare?
AI-901 is a Fundamentals exam, but 55–60% of it covers building AI solutions with Microsoft Foundry, and Microsoft expects basic Python knowledge. If you can write simple Python, about four weeks of study with hands-on Foundry practice and timed practice tests is a realistic plan.
Should I take AI-900 or AI-901?
Take AI-901. Microsoft’s AI-900 exam page says AI-900 retired on 30 June 2026 and was replaced by AI-901, which is now the only exam for the Azure AI Fundamentals certification. Certifications already earned with AI-900 stay valid.
Can I take the AI-901 exam online?
Yes. Microsoft’s registration page says that in most cases you can take an exam online with a proctor or at a test centre. If you choose online, run the Pearson VUE system pre-check on your computer before you register.
Does the Azure AI Fundamentals certification expire?
No. Microsoft’s certification renewal page says Fundamentals certifications do not expire, so there is no yearly renewal for Azure AI Fundamentals. Associate certifications such as the one earned with AI-103 must be renewed each year with a free online assessment.
Official exam page: Exam AI-901: Microsoft Azure AI Fundamentals on Microsoft Learn. Checked on 9 November 2026.
Start practising for AI-901
Practise with six 60-question tests mapped to the official AI-901 skill areas, including a timed Exam Mode test, with every answer explained.
✓ Every answer explained · ✓ Timed exam mode · ✓ 30-day refund