AI-300 is Microsoft's exam titled Operationalizing Machine Learning and Generative AI Solutions. Passing it earns the Microsoft Certified: Machine Learning Operations Engineer Associate certification. It is the replacement for DP-100, the Azure Data Scientist Associate, which retired on June 1, 2026.

Yes. DP-100 and the Azure Data Scientist Associate certification retired on June 1, 2026. Microsoft's certification page for it now carries a warning stating that this certification and the renewal assessment are retired. It was the first of the six certifications in Microsoft's Cloud and AI retirement wave to go.

What is the difference between AI-300 and DP-100?

The role changed from data scientist to operations engineer. DP-100 was about designing and running data science experiments. AI-300 is about operationalizing them: MLOps infrastructure, Bicep, GitHub Actions, model registries, drift detection, and an entirely new GenAIOps half covering Foundry environments, evaluation and observability for generative systems.

What are the AI-300 domain weights?

Five domains. Implement machine learning model lifecycle and operations is the largest at 25 to 30 percent. Design and implement a GenAIOps infrastructure is 20 to 25 percent. Design and implement an MLOps infrastructure is 15 to 20 percent. Generative AI quality assurance and observability, and optimizing generative AI systems, are each 10 to 15 percent.

Does AI-300 still cover classical machine learning?

Yes, and it is the only exam in the AI wave that does. The phrase machine learning appears 12 times in AI-300's outline and zero times in AI-901's. MLflow experiment tracking, automated machine learning, hyperparameter tuning, distributed training and data drift are all explicitly examined.

Is there an official AI-300 practice assessment?

No. Microsoft's certification page states the practice assessment is not currently available, and that practice assessments usually appear within 8 weeks of an exam leaving beta and becoming generally available. There is no official AI-300 question bank at any price.

What score do you need to pass AI-300?

700 or greater on a scale of 1 to 1000. Microsoft notes that because this is a scaled score it may not equal 70 percent of the points. Microsoft publishes no question count and no pass rate for AI-300 or any exam, so treat any exact figure you see as invented.

Does my DP-100 certification still count?

Yes, until it expires. Certifications already earned stay valid through their existing expiration date after the exam retires. What you lose is the renewal, because the renewal assessment retired with the exam. To stay current in this role you move to AI-300.

AI-300 practice test

AI-300 Practice Test, Practice Questions and Exam Prep From Your Own MLOps Notes

AI-300 replaced DP-100, and no official practice assessment exists for it. Upload your Azure Machine Learning notes, MLOps runbooks or prep PDF, and the AI writes unlimited AI-300 practice questions with an answer key in seconds.

Your study files are processed securely and deleted automatically after your practice questions are built.

Upload your MLOps notes or prep PDF and generate your first question set

Click to upload or drag and drop

PDF, DOCX, PPTX, TXT, JPG, JPEG, PNG, HEIC, ODP, ODT, BMP, or TIFF

up to 20MB

Please wait, your quiz is being created...

Uploading...

AI-300 is the exam that replaced DP-100, and it changed the job, not just the code. Its title is Operationalizing Machine Learning and Generative AI Solutions, and passing it earns the Machine Learning Operations Engineer Associate certification. DP-100 and the Azure Data Scientist Associate retired on June 1, 2026, the first casualty of Microsoft's Cloud and AI wave. The role moved from data scientist to operations engineer: you are no longer being tested on building the model, you are being tested on shipping, versioning, monitoring and governing it. Roughly 45 to 55 percent of the exam is generative AI operations, a topic DP-100 barely touched. Pass mark is 700 out of 1000 and there is no official practice assessment.

Last updated July 2026

Microsoft did not rename the data scientist exam. It replaced the role.

The clearest way to see what happened is to read the two certification titles next to each other. DP-100 earned you Azure Data Scientist Associate. AI-300 earns you Machine Learning Operations Engineer Associate. That is not a rebrand, that is a different person on the org chart.

DP-100 asked whether you could run a data science project on Azure: design an experiment, explore the data, train a model, tune it, explain it. AI-300 largely assumes the model exists, and asks whether you can put it into production and keep it alive. Bicep and Azure CLI to stand the workspace up. GitHub Actions to provision it. MLflow to track the experiments. Registries to share assets. Real-time and batch endpoints with progressive rollout and safe rollback. Drift detection with retraining triggers when thresholds are exceeded.

And then there is the half of the exam that did not exist before at all.

Passing score: 700 / 1000
Domains: 5
Generative AI ops: 45 to 55%
Practice questions: Unlimited

Half of AI-300 is GenAIOps, and it is entirely new material

Add up the three generative domains: GenAIOps infrastructure at 20 to 25 percent, generative AI quality assurance and observability at 10 to 15 percent, and optimizing generative AI systems at 10 to 15 percent. That is 40 to 55 percent of the exam, and almost none of it has a counterpart on DP-100.

This is the part experienced data scientists underestimate. You can have shipped models for a decade and still have never built an evaluation harness for a RAG system, never instrumented a generative app for groundedness, never had to explain why a fine-tune was the wrong call. Those are the questions.

Term in the official outline	AI-300	What it tells you
Machine learning	12	Classical ML survives here, and only here. AI-901 mentions it zero times.
Generative AI	6	A first-class citizen, not an appendix.
Fine-tuning	6	The most heavily named optimization technique on the exam.
Prompt	5	Prompt work is examined as an optimization lever, alongside fine-tuning.
Retrieval, RAG	5	You are expected to operate a RAG system, not just build one.
MLOps	2	Named explicitly. The whole exam is organized around it.

Note the fine-tuning count. Six mentions makes it the most named optimization technique in the outline, which is a genuine signal about where the questions concentrate. A lot of prep material treats fine-tuning as a footnote next to prompting and RAG. On this exam it is not.

Because there is no official bank, the reliable move is to build questions from the current documentation and your own pipelines. Upload your team's MLOps runbook and generate practice questions from your own deployment notes, which has the useful side effect of testing whether the runbook is actually correct.

AI-300 skills measured, straight from Microsoft's study guide

Five domains. The lifecycle domain is the biggest, and the three generative domains together outweigh it.

Domain	What is actually in it	Weight
Implement machine learning model lifecycle and operations	Orchestrating training: MLflow experiment tracking, automated machine learning, notebooks, hyperparameter tuning, distributed training for deep learning, training pipelines, comparing performance across jobs. Model registration and versioning, including packaging a feature retrieval specification with the artifact and evaluating a model against responsible AI principles. Deployment as real-time or batch endpoints with progressive rollout and safe rollback. Monitoring: detecting data drift, tracking production metrics, and configuring retraining or alert triggers. The largest domain.	25 to 30%
Design and implement a GenAIOps infrastructure	Creating and configuring Foundry environments and platform configuration, and the operational scaffolding around generative systems.	20 to 25%
Design and implement an MLOps infrastructure	Machine Learning workspaces, datastores, compute targets, identity and access management. Data assets, environments, components, and sharing assets across workspaces via registries. Infrastructure as code: GitHub integration, deploying workspaces with Bicep and Azure CLI, automating provisioning with GitHub Actions, restricting network access, and Git source control.	15 to 20%
Implement generative AI quality assurance and observability	Evaluating generative systems and instrumenting them so you can see what they are doing in production.	10 to 15%
Optimize generative AI systems and model performance	Fine-tuning, prompt-level optimization, and the performance and cost trade-offs between them.	10 to 15%

One structural point worth noticing. AI-300's study guide has an "About the exam" section rather than the "Updates to the exam" block Microsoft uses on established exams, and there is no dated "skills measured as of" version history. That is the fingerprint of an exam with no previous edition. There is no older outline to fall back on and nothing to cross-check against, which is precisely why recycled DP-100 material is worse than useless here.

Microsoft does not publish a question count for AI-300, and publishes no pass rate for any exam. Its general guidance is that most exams "typically contain between 40-60 questions." Microsoft's exam pricing feed does not yet list a US price for AI-300; associate role-based exams are normally $165 in the United States, but confirm at checkout rather than trusting a third-party figure.

How to build AI-300 practice questions with no official bank

A brand-new exam with no previous edition means every question bank on sale was written from guesswork.

1 Upload your pipelines: Azure ML documentation, your Bicep templates, your GitHub Actions workflows, your team's MLOps runbook.
2 Weight GenAIOps: The three generative domains are up to 55 percent combined and are new to almost everyone.
3 AI writes the questions: Exam-style items with an answer key and explanations, built from current material rather than a DP-100 bank.
4 Do not skip fine-tuning: It is the most-named optimization technique in the outline, and most courses treat it as an afterthought.

AI-300 questions, answered

What is Exam AI-300?: AI-300 is Microsoft's exam titled Operationalizing Machine Learning and Generative AI Solutions. Passing it earns the Microsoft Certified: Machine Learning Operations Engineer Associate certification. It is the replacement for DP-100, the Azure Data Scientist Associate, which retired on June 1, 2026.
Is DP-100 retired?: Yes. DP-100 and the Azure Data Scientist Associate certification retired on June 1, 2026. Microsoft's certification page for it now carries a warning stating that this certification and the renewal assessment are retired. It was the first of the six certifications in Microsoft's Cloud and AI retirement wave to go.
What is the difference between AI-300 and DP-100?: The role changed from data scientist to operations engineer. DP-100 was about designing and running data science experiments. AI-300 is about operationalizing them: MLOps infrastructure, Bicep, GitHub Actions, model registries, drift detection, and an entirely new GenAIOps half covering Foundry environments, evaluation and observability for generative systems.
What are the AI-300 domain weights?: Five domains. Implement machine learning model lifecycle and operations is the largest at 25 to 30 percent. Design and implement a GenAIOps infrastructure is 20 to 25 percent. Design and implement an MLOps infrastructure is 15 to 20 percent. Generative AI quality assurance and observability, and optimizing generative AI systems, are each 10 to 15 percent.
Does AI-300 still cover classical machine learning?: Yes, and it is the only exam in the AI wave that does. The phrase machine learning appears 12 times in AI-300's outline and zero times in AI-901's. MLflow experiment tracking, automated machine learning, hyperparameter tuning, distributed training and data drift are all explicitly examined.
Is there an official AI-300 practice assessment?: No. Microsoft's certification page states the practice assessment is not currently available, and that practice assessments usually appear within 8 weeks of an exam leaving beta and becoming generally available. There is no official AI-300 question bank at any price.
What score do you need to pass AI-300?: 700 or greater on a scale of 1 to 1000. Microsoft notes that because this is a scaled score it may not equal 70 percent of the points. Microsoft publishes no question count and no pass rate for AI-300 or any exam, so treat any exact figure you see as invented.
Does my DP-100 certification still count?: Yes, until it expires. Certifications already earned stay valid through their existing expiration date after the exam retires. What you lose is the renewal, because the renewal assessment retired with the exam. To stay current in this role you move to AI-300.

PDFQuiz is not affiliated with, endorsed by, or sponsored by Microsoft. Microsoft, Azure and Microsoft Foundry are trademarks of the Microsoft group of companies. This generator builds practice questions from material you upload and is a study aid, not a substitute for Microsoft's official learning paths. AI-300 is a new exam and its details are still settling: always confirm on Microsoft's own pages before you book.

Related study tools

The developer-side exam in the same wave is AI-103, which covers building AI apps and agents rather than operating them. The fundamentals exam that dropped all its machine learning content is AI-901, the replacement for AI-900. For the backend platform under an AI application, see AI-200. If you work on the data platform side, the PL-300 practice test generator covers Power BI, and infrastructure engineers usually pair this with AZ-104 and AZ-305. For any source document at all, use the certification exam generator.

Build your first AI-300 practice set

No official AI-300 bank exists. Upload your MLOps notes and generate exam-style questions in under a minute.

Start Creating Practice Questions

Z tej samej rodziny narzędzi