Microsoft · Exam reference
About the Microsoft AI-901 exam
Microsoft AI-901 is the exam behind the Azure AI Fundamentals certification, and it replaced the retired AI-900 on 30 June 2026. It is no longer a describe-only exam: more than half of it asks you to build working AI solutions in Microsoft Foundry, using Python and the Foundry SDK.
10
Sample questions
60 min
Exam time limit
70%
Practice pass mark
$99
Exam voucher
Microsoft AI-901, Azure AI Fundamentals, is the single exam that earns the Microsoft Certified: Azure AI Fundamentals credential. It took over from AI-900 when that exam retired on 30 June 2026, and the two are not the same test wearing a new number. AI-900 asked you to describe machine learning concepts and name Azure services. AI-901 splits into two halves: a concepts half worth 40 to 45 percent covering responsible AI, how generative models work, how to choose and configure a model, and how to recognise every workload type on the outline; and an implementation half worth 55 to 60 percent that lives entirely inside Microsoft Foundry. That second half is the one that catches people out. Microsoft's own audience profile now asks for knowledge of Python coding syntax and programming techniques, and the outline uses verbs like deploy, create and build rather than describe. You are expected to have deployed a model in the Foundry portal, written a lightweight chat client with the Foundry SDK, created and tested a single-agent solution, and pulled structured data out of documents, images and recordings with Content Understanding. The machine learning domain that dominated AI-900 is gone entirely. This exam suits anyone starting out in AI solution development: developers moving into AI, IT professionals whose organisation has adopted Azure AI, students, and career changers who want a credential that reflects how AI is actually built in 2026 rather than how it was taught in 2021. It is not a prerequisite for the associate-level Azure AI Engineer certification, but it is the natural first step towards it.
Exam Domains Covered
Exam Format & Details
Multiple-choice and multiple-response items delivered through Pearson VUE, online-proctored or at a test centre. Microsoft does not publish a fixed question count or duration for AI-901; expect roughly 40-60 questions in about 60 minutes of seat time. Scoring is scaled 1-1000 and you need 700 to pass, which is the 70% shown here. The voucher is USD 99 in the United States, with local pricing elsewhere. There are no hands-on labs or performance-based simulations, but a substantial share of the questions show you Python and Foundry SDK code and ask you to read it. The English version of the skills-measured outline was last updated on 15 April 2026.
Why Practice Questions Matter
Most AI-901 material on sale is recycled AI-900 material with new headings pasted over it, so it drills a machine learning domain the exam no longer contains and calls Foundry by names Microsoft has stopped using. Practice questions written against the current 15 April 2026 outline fix that. They also train the specific skill this exam rewards: reading a short block of Python or Foundry SDK code and deciding what it does, which is very different from recalling a definition. Working through a large bank domain by domain shows you quickly whether your weakness is the concepts half or the Foundry half, so your remaining study time goes where it actually counts.
Sample Practice Questions
The following questions are a preview of the type of syllabus-aligned questions you will practise in CertCrush. They reflect the format and reasoning style of the Microsoft AI-901 exam — not actual exam content.
Q1.Which command installs the Speech SDK used by the speech to text script in this chapter?
- A.`pip install azure-cognitiveservices-speech`
- B.`pip install azure-ai-projects`
- C.`pip install azure-speech-sdk`
- D.`pip install openai azure-identity`
Domain: Implement AI solutions by using Microsoft Foundry
Q2.A retailer's assistant is asked to write a launch email. Finding the product specification missing, it queries the catalogue application programming interface, checks stock levels, and keeps going until it judges it has enough detail, then writes the email. Which workload family is this?
- A.Agentic AI
- B.Generative AI, because the deliverable is written copy
- C.Information extraction, because it pulls product details
- D.Text analysis, because it reads the catalogue
Domain: Identify AI concepts and capabilities
Q3.An architect is listing everything that is counted or capped in tokens on a Foundry deployment. According to the guide, which of the following are measured in tokens? (Choose THREE.)
- A.Billing for a model call, with input and output counted separately
- B.The rate limit on a deployment, expressed per minute
- C.The context window available for a single call
- D.The number of concurrent users allowed on a deployment
- E.The Azure region a deployment is hosted in
- F.The number of documents stored in a search index
Domain: Identify AI concepts and capabilities
Q4.An editor complains the model keeps reusing the same words, while a second reviewer wants it nudged toward new topics. Which parameters address each, and what range do they share?
- A.Frequency penalty for repeated words and presence penalty for new topics, both -2.0 to 2.0
- B.Presence penalty for repeated words and frequency penalty for new topics, both 0 to 2
- C.Temperature for repeated words and top_p for new topics, both 0 to 2
- D.Seed for repeated words and streaming for new topics, both true or false switches
Domain: Identify AI concepts and capabilities
Q5.In the describe and redraw lab the vision call sets `{"type": "input_image", "image_url": SOURCE_URL, "detail": "high"}` and the following generation call sets `size="1024x1536"`. Why those two choices?
- A.High detail makes the model notice small features worth redrawing, and 1024x1536 gives the portrait shape a poster needs
- B.High detail reduces token cost, and 1024x1536 is the fastest size to generate
- C.High detail is required whenever image_url is a public web address, and 1024x1536 is the only supported size
- D.High detail forces the model to return a prompt under 40 words, and 1024x1536 is square
Domain: Implement AI solutions by using Microsoft Foundry
Frequently Asked Questions
What is included in the free Microsoft AI-901 sample?
The free sample includes 10 syllabus-aligned practice questions, sample flashcards, and a preview chapter from the study guide. No account or payment is required to try the sample.
How many questions are in the full Microsoft AI-901 course?
The full course includes a comprehensive question bank covering all exam domains. You can see the total question count on the Microsoft AI-901 course page.
Are these official Microsoft exam questions?
No. CertCrush questions are independently written and syllabus-aligned — they mirror the format, difficulty, and reasoning style of the official exam. We are not affiliated with or endorsed by Microsoft.
Which domains does the Microsoft AI-901 course cover?
The course covers 2 exam domains: Identify AI concepts and capabilities, Implement AI solutions by using Microsoft Foundry.
Can I study on mobile?
Yes. CertCrush is fully responsive and works on phones, tablets, and desktops. The timed exam, flashcards, and study guide all work on mobile without installing an app.
What happens when I create an account?
Creating a free account lets you access full courses, track your weak areas by domain, and resume practice sessions across devices. No credit card is required to register.