AI-901: Microsoft Azure AI Fundamentals replaced AI-900 when it retired on 30 June 2026, and it is still the exam for the Microsoft Certified: Azure AI Fundamentals certification. The name is the same, but the exam is very different. AI-900 was mostly concepts. In AI-901, more than half the marks (55–60%) are for building small AI solutions with Microsoft Foundry, and Microsoft expects you to know basic Python syntax.
This guide follows Microsoft's study guide (skills measured as of 15 April 2026). It covers both skill areas, the hands-on tasks you must be able to do, a 3-week plan and 20 original practice questions.
Quick facts (Microsoft Learn, November 2026)
| Item | Detail |
|---|---|
| Exam | AI-901: Microsoft Azure AI Fundamentals |
| Certification | Microsoft Certified: Azure AI Fundamentals |
| Level | Fundamentals, no prerequisites |
| Passing score | 700 |
| Expiry | Fundamentals certifications don't expire |
| Language for code | Basic Python |
| Price | Set by the country where you take the exam |
The free Practice Assessment is on AI Skills Navigator. Read the official AI-901 study guide before you book.
AI-900 vs AI-901: what changed?
| AI-900 (retired) | AI-901 (current) | |
|---|---|---|
| Style | Mostly conceptual | Concepts plus hands-on Foundry tasks |
| Coding | Not required | Basic Python syntax expected |
| Focus | ML, vision, NLP, generative AI concepts | Generative AI and agents, text, speech, vision, information extraction in Foundry |
| Weights | Several concept domains | Concepts 40–45%, implementation 55–60% |
If you already passed AI-900, it stays valid, because fundamentals certifications don't expire. For all the 2026 changes, see Microsoft certifications retired in 2026.
Who should take AI-901?
- Freshers and students starting in AI.
- Developers, testers and analysts who want a first Azure AI credential.
- Sales and pre-sales people who need hands-on AI vocabulary.
AI-901 pairs well with AZ-900: Azure basics plus AI basics. After it, the natural next step is AI-103.
Skill area 1: Identify AI concepts and capabilities (40–45%)
Responsible AI principles. Know each one with an example:
| Principle | Example question to ask |
|---|---|
| Fairness | Does a loan model treat similar applicants differently by gender? |
| Reliability and safety | Does the system behave safely when inputs are unusual? |
| Privacy and security | Is personal data protected and access controlled? |
| Inclusiveness | Can people with disabilities use it (e.g., alt text, speech)? |
| Transparency | Do users know they are talking to AI and how it decides? |
| Accountability | Who is responsible when it goes wrong? |
Models:
- how generative models work: tokens, prediction and context;
- choosing a model by capability (text, multimodal, small vs large);
- deployment options and settings such as temperature and maximum tokens.
AI workloads:
- recognise generative and agentic AI, text analysis, speech, computer vision and image generation, and information extraction;
- text techniques: key-phrase extraction, entity detection, sentiment analysis and summarisation;
- speech recognition and speech synthesis;
- extracting information from text, images, audio and video.
Skill area 2: Implement AI solutions with Microsoft Foundry (55–60%)
You should be able to do each of these yourself, not just describe it:
Generative AI apps and agents:
- write effective system and user prompts;
- deploy a model and chat with it in the Foundry portal;
- build a small chat client with the Foundry SDK;
- create and test a single agent in the portal;
- build a small client application for that agent.
Text and speech:
- a small app that includes text analysis;
- responding to spoken prompts with a multimodal model;
- a small app using Azure Speech in Foundry Tools.
Vision and image generation:
- interpret images in prompts with a multimodal model;
- generate new images;
- a small app with vision features.
Information extraction with Azure Content Understanding:
- extract information from documents and forms, images, audio and video;
- a small app that uses those extractions.
Minimal chat client (Python)
# pip install azure-ai-projects azure-identity
from azure.identity import DefaultAzureCredential
from azure.ai.projects import AIProjectClient
project = AIProjectClient(
endpoint="https://<resource>.services.ai.azure.com/api/projects/<project>",
credential=DefaultAzureCredential(), # sign in with az login first
)
client = project.get_openai_client()
history = "You are a helpful assistant for a Hyderabad IT training institute."
while True:
q = input("You: ")
if q.lower() in {"exit", "quit"}:
break
r = client.responses.create(model="<deployment-name>", instructions=history, input=q)
print("AI:", r.output_text)That covers "create a lightweight chat client by using the Foundry SDK". Practise extending it, for example by adding an image to the prompt or calling an agent instead of a model. The Microsoft Foundry tutorial walks through the platform.
3-week study plan
| Days | Focus | Hands-on |
|---|---|---|
| 1–3 | Responsible AI, how generative AI works, model choice | Write 6 principle examples from your own life |
| 4–6 | Foundry portal: deploy a model, chat playground, prompts | Compare temperature 0.2 vs 0.9 on the same prompt |
| 7–9 | Python SDK chat client, single agent in the portal | The client above, then an agent with instructions |
| 10–12 | Text analysis and speech | App that summarises text and reads the summary aloud |
| 13–15 | Vision and image generation | Describe a photo; generate a poster image |
| 16–18 | Content Understanding: documents, images, audio, video | Extract fields from a sample invoice |
| 19–21 | Practice Assessment twice, revise weak areas | Book the exam |
Use a free or trial Azure subscription and delete resources after each lab.
20 practice questions with answers
These are original practice questions written for this guide, not exam questions.
1. A hiring model rejects more women with equal qualifications. Which principle is violated? Fairness.
2. Users must be told they are chatting with an AI. Which principle? Transparency.
3. A medical-advice bot gives dangerous answers for unusual inputs. Which principle? Reliability and safety.
4. You want more consistent, less creative answers. Which setting do you lower? Temperature.
5. The task is answering questions about a photo. Which kind of model do you need? A multimodal model.
6. Pull out names, places and dates from news text. Which technique? Entity detection (entity recognition).
7. Decide whether reviews are positive or negative. Which technique? Sentiment analysis.
8. Turn a meeting recording into text. Which capability? Speech recognition (speech-to-text).
9. Read an answer aloud to the user. Which capability? Speech synthesis (text-to-speech).
10. Extract invoice number, date and total from scanned invoices. Which Foundry tool? Azure Content Understanding.
11. Where do you first try a deployed model without writing code? The chat playground in the Foundry portal.
12. You need a small Python program that chats with your deployed model. What do you use? The Foundry SDK (project client), with a chat client built on it.
13. An assistant should follow fixed rules, such as "only answer about our courses". Where do you put them? In the system prompt (the instructions).
14. An AI that can call tools and complete multi-step tasks for a user is called what? An agent.
15. Create a new poster image from a text description. Which capability? Image generation.
16. Extract the spoken key points from a training video. Which tool handles audio and video? Content Understanding, which works on audio and video as well as documents and images.
17. Shorten a long policy document into key points. Which technique? Summarisation.
18. An app must work for people with low vision. Which principle, and what feature helps? Inclusiveness. Speech output and good alt text help.
19. Who should answer for an AI system's mistakes? Accountability: named people and processes within the organisation.
20. Is AI-900 still valid if you passed it in 2025? Yes. Fundamentals certifications don't expire, but new candidates now take AI-901.
After AI-901
- Developer path: AI-103 (AI apps and agents) or AI-200 (AI cloud developer).
- Administrator path: AZ-104.
- Careers: read FDE vs AI Engineer vs GenAI Engineer and see AI jobs for freshers.
Cloud Soft Solutions in Ameerpet, Hyderabad runs Azure AI and generative AI training with hands-on labs. See all courses and the APEX program.
📞 Call +91 96660 19191 · WhatsApp · Flat No. 513, 5th Floor, Nilagiri Block, Aditya Enclave, Ameerpet, Hyderabad
Frequently Asked Questions
Is AI-900 still available?
No. AI-900 retired on 30 June 2026 and was replaced by AI-901, which keeps the Azure AI Fundamentals certification name.
Is AI-901 harder than AI-900?
It is more hands-on: 55–60% covers implementing solutions with Microsoft Foundry, and basic Python syntax is expected.
What is the AI-901 passing score?
700.
Does the Azure AI Fundamentals certification expire?
No. Microsoft fundamentals certifications don't expire.
Is my old AI-900 still valid?
Yes. Fundamentals certifications don't expire, so an AI-900 you earned stays on your profile.
What should I take after AI-901?
AI-103 for AI apps and agents, AI-200 for back-end AI development, or AZ-104 for Azure administration.
