Microsoft Azure AI Fundamentals: the honest guide
AI-901 is not AI-900 with a new number on it. Microsoft updated the exam on 15 April 2026 and moved the centre of gravity: 57 percent of the questions now cover implementing AI solutions with Microsoft Foundry, and the audience profile asks for knowledge of Python coding syntax and programming techniques along with familiarity with Azure resources. The exam it replaced was a vocabulary test you could pass by reading. This one is not, and preparing as though it were is the most expensive mistake available on this page.
The format is 40 to 60 questions in 45 minutes, using multiple choice, drag-and-drop, and hot area items where you click a region of a screenshot. You pass at 700 on a scale of 1 to 1000. That is a scaled score, not a percentage, so 700 does not mean you got 70 percent of the questions right.
The other 43 percent is identifying AI concepts and capabilities: the responsible AI principles, how generative models work, choosing a model and its deployment options, and recognising generative, text, speech, vision and information extraction workloads. That domain rewards reading. The larger one rewards having built something.
Who Azure AI Fundamentals is for
A good fit if
- You read Python, or can learn to in a fortnight, and your team is starting to put model calls into an application.
- You already work in Azure and want a cheap credential that evidences deploying a model rather than defining one.
- You are heading toward AI-102 and want a graded checkpoint before committing to a role-based exam.
- Your employer is putting chat clients and agents into production and you need the vocabulary and the portal at the same time.
- You are willing to build the exam's example solutions on an Azure free account instead of reading about them.
Probably not, if
- You wanted the exam AI-900 used to be. A credential you could earn from a weekend of reading definitions is not what this is any more, and preparing that way is the commonest reason candidates fail it.
- You are non-technical and want shared cloud vocabulary for a migration or a procurement conversation. AZ-900 costs the same, has no code in it, and fits that goal better.
- You cannot read a short Python script and do not intend to learn. Microsoft states the expectation in the audience profile and the implementation domain assumes it throughout.
- You already build with the Foundry SDK in your day job. AI-102 is the exam that matches what you can already do; this one will tell a hiring manager less than your work does.
Is Azure AI Fundamentals worth it?
For someone working near AI in Azure, yes, and more so than before the update. At $99 with no expiry it was already the cheapest Microsoft credential worth holding. The April 2026 rewrite gave it content: passing now says you deployed a model in the Foundry portal, called it from a client, and pulled information out of a document, rather than that you could define machine learning.
For a non-technical reader the answer moved the other way, and it is worth saying plainly. The implementation domain is 57 percent of the questions and it is not answerable from a glossary. If the goal is holding your own in a meeting about an AI project, AZ-900 is the same price and asks nothing of you in Python.
For someone already building AI solutions on Azure, no. This is a fundamentals exam and hiring managers read it as one. AI-102 is the credential that changes how a CV is read, and AI-901 is not a prerequisite for it.
For someone sitting on AI-900 study material they never used, roughly half of it is still good. The concepts domain survives the update in shape. The rest of those notes describe an exam that no longer exists, and the honest move is to throw them out rather than to hope the overlap carries you.
What the exam actually asks you to do
Multiple choice and multiple response, plus drag-and-drop ordering and hot area items where you click a region of a screenshot.
- Multiple choice
- Multiple response
- Drag and drop
- Hot area
The highlighted formats are the ones you cannot answer from memory alone. Microsoft Learn, exam duration and question types ↗
Domain breakdown and official weightings
From the official Microsoft Learn study guides. Implement AI solutions by using Microsoft Foundry is the heaviest domain at 57 percent, followed by Identify AI concepts and capabilities at 43 percent.
- Identify AI concepts and capabilities43%
- Implement AI solutions by using Microsoft Foundry57%
Study plans by experience level
Comfortable with Python, already working in Azure
2 to 3 weeks at 6 to 8 hours
- 1Read the April 2026 skills outline first, and treat the second domain as a build list rather than a reading list. Every bullet under it starts with a verb: deploy, create, build, extract.
- 2Session one: create an Azure free account, deploy a model in the Foundry portal, and interact with it there. Write the system prompt and the user prompt yourself, then change them until the output changes the way you predicted.
- 3Session two: build the lightweight chat client with the Foundry SDK, then create and test a single-agent solution in the portal and call it from a client. These are four separate bullets in the outline and the densest part of the exam.
- 4Session three: text and speech. Build something small that runs text analysis, then respond to a spoken prompt using a deployed multimodal model and Azure Speech in Foundry Tools.
- 5Session four: vision and information extraction. Put an image into a prompt for a multimodal model, generate an image, then run a document, an image and an audio file through Content Understanding.
- 6Final week: the concepts domain, which is 43 percent and mostly reading. Learn the responsible AI principles by name, with one consideration you can state for each: fairness, reliability and safety, privacy and security, inclusiveness, transparency, accountability.
- 7Sit one timed set at 45 minutes before booking. The clock is the second failure mode on this exam and it is free to test for.
Working in IT, new to Python
6 to 8 weeks at 5 to 6 hours
- 1Weeks 1 to 2: Python for reading, not for writing from scratch. Variables, functions, lists and dictionaries, importing a library, creating a client object, and reading a JSON response. The exam shows you code far more often than it asks you to produce it.
- 2Week 3: Azure resources. Resource groups, regions, endpoints and keys, and where a deployment actually lives. The audience profile names this and the Foundry work assumes it silently.
- 3Weeks 4 to 6: work the Foundry domain one sub-objective at a time, building each one before moving on. Model deployment and prompts, the SDK chat client, a single agent, text analysis and speech, vision and image generation, then Content Understanding.
- 4Week 7: the concepts domain. Responsible AI, how generative models work, picking a model and a deployment configuration, and the workload types you are expected to recognise.
- 5Week 8: timed practice at 45 minutes, plus enough drag-and-drop and hot area items that the interaction is unremarkable rather than novel.
Converting from AI-900 study material
3 to 4 weeks at 6 hours
- 1Put the old notes next to the current skills outline and mark what still applies. Responsible AI and workload identification survive. Material built around touring individual services, or around the exam being a vocabulary test, does not.
- 2Accept the reweighting before planning anything else. 57 percent of the questions are implementation, so more than half your remaining study time belongs in the Foundry portal even if none of your previous study time did.
- 3Week 1: a Python syntax check. If you can read a script that creates a client, sends a prompt and prints the response, you are fine. If you cannot, this is where the fourth week goes.
- 4Weeks 2 to 3: build every bullet in the Foundry domain once. Deploy and prompt a model, write the chat client, create an agent, run text analysis, use Azure Speech, prompt with an image, generate an image, and extract from a document, an image and an audio file with Content Understanding.
- 5Week 4: revisit the concepts domain, which is where your old material is strongest, then sit a timed set. Book only after the implementation questions stop feeling like guesses.
Non-technical, and the certification is required of you
8 to 10 weeks at 4 to 5 hours
- 1Be honest about the size of this before you start. The exam expects Python syntax and Azure familiarity, and at 57 percent implementation weighting neither is optional. If nobody is requiring AI-901 by name, AZ-900 is the same $99 with no code in it.
- 2Weeks 1 to 3: Python syntax, to reading level only. Type the examples out in a notebook rather than watching someone else type them.
- 3Weeks 4 to 5: an Azure free account, and what a resource group, a region and an endpoint are. Deploy one small thing and delete it.
- 4Weeks 6 to 9: the Foundry domain, one bullet per session, typing the samples rather than copying them. Working out what a piece of code returns is most of what the implementation questions ask.
- 5Week 10: the concepts and responsible AI domain, which is the part your existing work already prepares you for, then timed practice at 45 minutes.
Common mistakes
Preparing for AI-900 by another name
The certification name and the level did not change on 15 April 2026, so most free material online still describes the old exam. AI-901 puts 57 percent of the questions on implementing solutions in Microsoft Foundry. Check the date on anything you study from before you trust it.
Reading about Microsoft Foundry instead of opening it
The implementation sub-objectives are written as tasks: deploy a model, create a chat client, test an agent, extract information from a document. Questions written from task bullets ask what happens when you perform the task. An hour in the portal is worth an evening of video here.
Skipping Python because this is a fundamentals exam
The audience profile asks for knowledge of Python coding syntax and programming techniques. You are not expected to write applications, but you are expected to look at a short script and say what it sends to the model and what it gets back.
Reading 700 as 70 percent
The pass mark is 700 on a scale of 1 to 1000, and it is a scaled score. Item difficulty and item type both feed into it, so no fixed number of correct answers maps to the line. Prepare to be comfortable rather than to be one mark over.
Meeting drag-and-drop and hot area items on exam day
Two of the three item types are not multiple choice. Drag-and-drop asks you to order or match, and hot area asks you to click a region of a screenshot. Both take longer than they look inside 45 minutes, and neither rewards a first encounter under pressure.
Treating the responsible AI principles as filler
They sit inside the 43 percent concepts domain, they are enumerated in the skills outline, and they are the most predictable marks on the paper. Fairness, reliability and safety, privacy and security, inclusiveness, transparency and accountability, each with a consideration you can state in a sentence.
Leaving a deployment running after a study session
The Azure free account covers this syllabus comfortably if you delete the resource group when you finish. The candidates who get an unexpected bill are the ones who deployed a model in week one and found it still running in week six.
What comes after passing
Fundamentals certifications do not expire, so there is nothing to renew, no continuing education requirement and no fee. The credential stays on your Microsoft transcript and you can stop thinking about it.
AI-102 is the named next step and it is a real step up: role-based rather than fundamentals, and it assumes you build these solutions rather than that you built each one once during exam prep. If AI-901 felt comfortable, spend a few months doing the work before booking it. Role-based certifications renew through a free online assessment on Microsoft Learn, so the ongoing cost stays time rather than money.
Keep what you built. The chat client, the agent and the Content Understanding script are worth more in an interview than the certificate is, because they are the part nobody can hand you. Push them somewhere you can point at, with a short note on what each one does.
If the goal was getting into AI work rather than documenting work you already do, treat this as the first of two. AI-901 gets your CV read. It does not get you hired on its own, and the update that made it harder did not change that.
Where people go next
Costs across the full renewal cycle are on the Azure AI Fundamentals cost page.
Frequently asked questions
Is AI-901 the same as AI-900?
No. AI-901 replaced AI-900 on 15 April 2026. It keeps the Azure AI Fundamentals name, but it reweights the exam toward building: 55 to 60 percent covers implementing AI solutions with Microsoft Foundry, and Microsoft now expects familiarity with Python syntax and Azure resources.
Do I need to know Python for AI-901?
You need to read it rather than write it. The audience profile asks for knowledge of Python coding syntax and programming techniques. Working out what a short script sends to a model and what it gets back is the level the exam actually tests.
How hard is AI-901?
Harder than the exam it replaced, and still a fundamentals exam. The concepts domain at 43 percent is readable. The implementation domain at 57 percent is only comfortable if you have deployed a model in the Foundry portal and called it yourself at least once.
What score do I need to pass AI-901?
700 on a scale of 1 to 1000. It is a scaled score rather than a percentage, so it does not correspond to a fixed number of correct answers. The exam runs 40 to 60 questions in 45 minutes across multiple choice, drag-and-drop and hot area items.
Does AI-901 expire?
Fundamentals certifications do not expire, so there is no renewal cost at all and no continuing education requirement. Role-based certifications such as AI-102 renew through a free online assessment on Microsoft Learn, which costs time rather than money.
Should I take AI-901 or AZ-900 first?
Either works, and they share the fundamentals price. Take AI-901 first if your work is near AI applications and you are comfortable reading Python. Take AZ-900 first if you want general cloud grounding with no code in the exam.
How long does it take to study for AI-901?
Two to three weeks at 6 to 8 hours if you write Python and already work in Azure. Six to eight weeks if Python is new to you. Three to four weeks if you are converting AI-900 material, most of which goes into the Foundry portal rather than into reading.
Can I still use AI-900 study material?
Only the concepts half, and only with the outline open beside it. Responsible AI and workload identification still apply. Anything treating the exam as a vocabulary test, or built around a tour of individual services, describes an exam that was replaced on 15 April 2026.
Keep reading
Azure AI Fundamentals practice questions
Free sample questions with the full explanation on every answer.
Azure AI Fundamentals passing score
The exact cut score, what kind of number it is, and the retake terms.
How hard is Azure AI Fundamentals?
An honest difficulty read from the format, the clock and the weights.
What Azure AI Fundamentals costs
The voucher price, the retake, and what renewal costs across the cycle.