AI‑102 certification review (2026): retired, and what AI‑103 changes
Microsoft retired the AI-102 exam on June 30, 2026. Here's what that means if you hold it, and whether its successor, AI-103, is worth taking.

Our verdict on Microsoft AI-102 (Azure AI Engineer Associate), retired; successor AI-103
AI-102 is retired, so the real choice is AI-103. Python developers building RAG apps and agents on Azure Foundry should take it; people outside the Azure world or focused on MLOps should look elsewhere.
Pricing AI-102 can no longer be booked. AI-103 costs about $165 in the US (EUR 126 in Belgium); retakes full price; Microsoft Learn course, practice assessment and renewal free
Visit Microsoft AI-102 (Azure AI Engineer Associate), retired; successor AI-103Here's the short version: you can't take AI-102 any more. Microsoft retired the exam, and the Azure AI Engineer Associate certification it led to, on June 30, 2026 at 11:59 PM Central time. Its replacement is AI-103, "Developing AI Apps and Agents on Azure", which earns a new credential called Azure AI Apps and Agents Developer Associate.
So this review is a bit unusual. It covers what the AI-102 certification was, what happens if you already hold it, and whether AI-103 is worth your money and evenings in late 2026. Nobody here sat either exam for this piece. We compared the two using Microsoft Learn's study guides and policy pages, independent prep sites and posts from candidates, as of early November 2026.
My view, up front: if you build AI apps on Azure for a living, AI-103 is probably the better certification of the two. It tests what Microsoft actually wants you building now (agents, retrieval and Foundry projects) instead of a tour of older service APIs.
AI-102 in brief
AI-102 was the exam code; "Designing and Implementing a Microsoft Azure AI Solution" was its full name. Passing it made you a Microsoft Certified: Azure AI Engineer Associate. The audience was developers using Python or C# who could build image, video, language, knowledge mining and generative AI solutions with REST APIs and SDKs, going by the certification page, which now carries a banner saying the certification and its renewal assessment are retired.
Here's how the final version stacked up against its successor.
| Detail | AI-102 (retired) | AI-103 (current) |
|---|---|---|
| Full name | Designing and Implementing a Microsoft Azure AI Solution | Developing AI Apps and Agents on Azure |
| Credential | Azure AI Engineer Associate | Azure AI Apps and Agents Developer Associate |
| Status | Retired June 30, 2026 | Bookable through Pearson VUE |
| Languages expected | Python or C# | Python |
| Exam time | 100 minutes (per OpenExamPrep) | 120 minutes |
| Passing score | 700 out of 1,000 | 700 out of 1,000 |
| Generative AI and agents weight | 15-20% plus 5-10% | 30-35% |
The AI-103 figures come from its certification page and study guide; the AI-102 exam length is from OpenExamPrep's comparison, since Microsoft's retired page no longer shows it.
One search quirk, while we're here. "AI" is also Air India's airline code, so if you came looking for AI 102 flight status today, you want the airline's own site, not this page.
Topics the exam covered
The last AI-102 outline is dated December 23, 2025, and it's still readable on the archived study guide. It had six domains. I've grouped them below with the AI-103 equivalent, because that's the comparison most people care about now.
Planning and managing an Azure AI solution
This was 20-25% of AI-102: picking the right Foundry service, creating resources, managing keys and authentication, monitoring costs and deploying containers. On AI-103 it grows to 25-30% and adds quotas, rate limits, drift monitoring, managed identity, private networking and "keyless credentials". That last change is telling. Microsoft clearly wants you to stop pasting account keys into config files (which, to be fair, plenty of tutorials still do).
Generative AI and agents
AI-102 split these into two domains: generative AI at 15-20% and agentic solutions at just 5-10%. You'd deploy Azure OpenAI models, build prompt flows, ground a model with RAG, generate images with DALL-E and fine-tune. Agents meant the Foundry Agent Service and Microsoft Agent Framework.
AI-103 merges both into one domain worth 30-35%, its biggest. The bullets get specific: tool schemas, function calling, conversation memory, multi-agent orchestration, tracing, token analytics and "self-critique loops". If you only study one area, it should be this one.
Computer vision
Both exams weight vision at 10-15%, but the content moved. AI-102 tested image analysis, OCR, handwriting, custom vision models and Video Indexer. AI-103 swaps much of that for image and video generation, inpainting, captions, visual question answering and Content Understanding pipelines, plus defending against prompt injection hidden as text inside images. That's a real-world threat, and I'm glad it's on a syllabus.
Language, speech and information extraction
Natural language processing was 15-20% on AI-102, covering key phrases, sentiment, PII detection, translation and speech. On AI-103 it shrinks to "text analysis" at 10-15% and leans on generative prompting with structured JSON outputs. Knowledge mining (15-20%) became information extraction (10-15%): indexing documents, images, audio and video; setting up semantic, hybrid and vector search; and feeding the results to agents.
Exam cost and renewal
Microsoft doesn't print one global price. The exam page says the price is "based on the country or region in which the exam is proctored", and it only appears once the page knows where you are. Here's what I could confirm.
| Item | Cost | Source |
|---|---|---|
| AI-103 exam, United States | About $165 (typical associate price) | MSCertQuiz, citing Microsoft's FAQ |
| AI-103 exam, Belgium | EUR 126 | Global Knowledge |
| Retake after a fail | Full fee again | MSCertQuiz |
| Microsoft Learn course and practice assessment | Free | Microsoft Learn |
| Annual renewal | Free | Microsoft renewal page |
Prices checked October 2026. Renewal works the same way it did for AI-102: associate certifications expire every year, you get a six-month window to pass a free, unproctored, open-book online assessment, and passing adds a year from the expiry date.
The beta deal's gone. Microsoft's launch post offered 80% off to the first 300 people who sat the AI-103 beta on or before May 7, 2026, and that window closed months ago.
If you already hold the AI-102 certification
You keep it. Going by Microsoft's retirement rules, a certification earned or renewed before retirement stays in the Active section of your transcript until it expires, then moves to Historical. What you can't do is renew it, because the renewal assessment retired too. There's no automatic upgrade either: Vlad Catrinescu's breakdown puts it bluntly as "no automatic migration, no free upgrade, no voucher, and no discount".
So, practically, your AI-102 badge has at most about a year of active life left. If your employer or a Microsoft partner requirement counts it, I'd plan for AI-103 well before that date. Partners do get a grace period: Microsoft's exam retirement page says retired certifications keep earning partner credit for one year after the retirement date.
Difficulty and what candidates reported
Hard numbers on pass rates don't exist publicly; Microsoft doesn't publish them. What I can tell you is how the scoring works. Microsoft's scoring page says 700 is a pass on a 1-1,000 scale, and because it's scaled, it "may not equal 70% of the points". OpenExamPrep's AI-103 fact sheet estimates 40-60 questions, though that's its own estimate and not a Microsoft figure.
The retirement caught some people mid-study. One learner on the KodeKloud forum wrote in June: "I started preparing for AI-102 then realized that it is expiring this month end." On Microsoft Q&A, Julian Sharp answered a similar question in April: "The new certification that replaces AI-102, is AI-103 and will be launched in beta later this April". Microsoft had said in its launch post to expect general availability in June 2026, and OpenExamPrep reported on October 26 that the exam page no longer shows a beta label.
I couldn't find enough recent first-hand AI-103 reports to say whether it's harder than AI-102. My guess is it feels harder for people coming from classic Cognitive Services work and easier for people who already build RAG apps and agents in Python. The extra 20 minutes suggests Microsoft expects longer scenario questions, but that's an inference, not something Microsoft has said.
A worked study plan with numbers
Say you've got 60 hours to prepare. Splitting them by the midpoint of each AI-103 weight gives you roughly this:
- Generative AI and agents (32.5%): about 20 hours.
- Planning, security and monitoring (27.5%): about 17 hours.
- Computer vision, text analysis and information extraction (12.5% each): about 8 hours apiece.
On the old AI-102 weights, generative AI plus agents averaged 25%, or about 15 hours of that same budget. A third more time goes to agents and generative work on AI-103, mostly taken from classic language features and knowledge mining.
Old AI-102 prep that breaks on AI-103
Plenty of AI-102 material is still floating around, and some of it'll mislead you. The weak spots, going by the two study guides:
- C# notes: AI-103's audience profile mentions Python only.
- CLU intents and custom question answering: these had real weight on AI-102 and don't appear by name in the AI-103 outline.
- Account-key authentication: AI-103 asks about managed identity and keyless access instead.
- Agents treated as a side topic: they're now the core of the largest domain.
Search, document extraction, content filters and prompt shields carry over well. Keep those labs.
Best suited to
AI-103 fits developers who already write Python and want proof they can ship Foundry-based apps: RAG systems, agents with tools, multimodal pipelines. It's an associate-level exam, so I don't think it's a sensible first certification if you've never touched Azure. If you're weighing whether a cert is worth it at all, our guide to AI engineering salaries and our explainer on AI engineer courses are better starting points.
Skip it if your day job is mostly data science or MLOps (there's a separate exam for that, below), or if your company runs on AWS or Google Cloud. A Microsoft-specific badge won't count for much there.
AI-102 alternatives in 2026
AI-103: Azure AI Apps and Agents Developer Associate
This is the direct successor and the obvious pick for most former AI-102 candidates. The free self-paced Course AI-103T00 covers generative AI apps, agents, knowledge connections and multimodal content, and the official practice assessment now lives on Microsoft's AI Skills Navigator.
AI-200: Azure AI Cloud Developer Associate
According to Vlad Catrinescu, AI-200 is the suggested replacement for AZ-204, the general Azure Developer exam that retired July 31, 2026. It suits you if you're more of a cloud app developer who adds AI features than a specialist building agents.
AI-300: Machine Learning Operations Engineer Associate
The same source lists AI-300 as the successor to DP-100, the data scientist exam that retired June 1, 2026. It leans on automation, CI/CD, drift detection and cost control. Pick it if you train and run models rather than wire them into apps. For broader, vendor-neutral learning, see our guides to AI training courses and free AI courses.
Verdict
AI-102 had a long run, and it's over. If you hold it, enjoy the remaining months and don't panic; it stays on your transcript, moving to Historical once it expires. If you were planning to take it, switch to AI-103 now and don't waste weeks on old outlines.
Honestly, I think the change is overdue. AI-102's last outline still gave agents only 5-10% even though Microsoft's own Foundry Agent Service and Agent Framework were already on that outline. AI-103 isn't cheap at around $165 a sitting, and the fact that retakes cost full price stings. But for a Python developer building on Foundry, it's probably the most relevant Microsoft AI certification you can sit in late 2026.
FAQ
Is AI-102 retired?
Yes. Microsoft retired Exam AI-102 and the Azure AI Engineer Associate certification on June 30, 2026, at 11:59 PM Central time. You can't book it, and the renewal assessment is retired too. If you already earned it, it stays active on your transcript until its normal expiry date, then moves to Historical.
What replaced the AI-102 certification?
Exam AI-103, "Developing AI Apps and Agents on Azure", which earns Microsoft Certified: Azure AI Apps and Agents Developer Associate. It's a separate credential, not a rename. You'll need to pass AI-103 in full; holding AI-102 gives you no discount or shortcut. Its audience profile also names Python only.
How much does the AI-103 certification cost?
Microsoft prices exams by country. In the United States associate exams typically cost $165, and Global Knowledge lists AI-103 at EUR 126 in Belgium. Retakes cost the full fee again. Microsoft's course, practice assessment and yearly renewal assessment are free.
How hard is AI-103 compared with AI-102?
Nobody can say for sure, since Microsoft doesn't publish pass rates. With 30-35% of the weight on generative AI and agents, I'd guess it feels harder if you mainly know classic vision and language APIs.
Can I still renew my Azure AI Engineer Associate certification?
No. The renewal assessment retired with the exam on June 30, 2026. Your certification stays active until its current expiry date; after that, you'd need to pass AI-103.
What we like
- Existing AI-102 holders keep the credential on their transcript until it expires
- AI-103 puts 30-35% of the weight on generative AI and agents, matching current Azure work
- Free Microsoft Learn course, practice assessment and annual renewal
- Clear published skills outline with domain weights
What could be better
- AI-102 retired June 30, 2026 and can't be taken or renewed
- No automatic migration, voucher or discount to AI-103
- Retakes cost the full exam fee
- C# developers lose out: AI-103's audience profile names Python only



