MicrosoftAvailable
AI-900
DP-100 Exam Guide: Retired Exam, Skills and Next Steps
Direct answer: Microsoft's study guide says Exam DP-100 was retired on June 1, 2026, and the certification page says the certification and its renewal assessment are retired. Do not buy a voucher to sit DP-100. The skills it measured (Azure Machine Learning, MLflow and language model optimization) are still a useful learning map.
Microsoft retired Exam DP-100. Its study guide says the exam was retired on June 1, 2026, at 11:59 PM Central Standard Time. The certification page says the certification and the renewal assessment are retired.
So do not buy a voucher to sit DP-100. Use this page as a map of the skills the exam measured and as a plan for learning Azure Machine Learning.
What the DP-100 proved and who it was built for
The credential is called Microsoft Certified: Azure Data Scientist Associate. Microsoft lists it as Intermediate level, for the Data Scientist role, in the Azure product family.
It showed you could apply data science and machine learning to run ML workloads on Azure. Microsoft also expected knowledge of optimizing language models for AI applications.
Microsoft lists these role responsibilities:
Designing and creating a working environment for data science workloads.
Exploring data and training machine learning models.
Implementing pipelines and running jobs to prepare for production.
Managing, deploying and monitoring scalable ML solutions.
Using language models to build AI applications.
Candidates were expected to have experience with Azure Machine Learning, MLflow, Azure AI services (including Azure AI Search) and Azure AI Foundry. Microsoft notes that Azure AI Foundry is now Microsoft Foundry and that it was updating the exam materials. The certification page also names Python as part of the work.
Skills measured and their weights
These are the skills measured as of April 11, 2025, from the official study guide.
Skill area | Weight |
|---|---|
Design and prepare a machine learning solution | 20–25% |
Explore data, and run experiments | 20–25% |
Train and deploy models | 25–30% |
Optimize language models for AI applications | 25–30% |
Two areas carried the most weight, and they pull in different directions. "Train and deploy models" is classic MLOps work. "Optimize language models" is newer and covers generative AI.
What sat inside each area
Design and prepare: choosing compute, creating workspaces, datastores, compute targets, data assets and environments, Git integration, and sharing assets through registries.
Explore data and run experiments: automated ML for tabular, computer vision and NLP data, notebooks, feature stores, MLflow tracking and hyperparameter tuning (sampling method, search space, primary metric, early termination).
Train and deploy: training scripts as jobs, pipelines with custom components, registering MLflow models, online endpoints and batch endpoints.
Optimize language models: the model catalog, prompt engineering and prompt flow, Retrieval Augmented Generation (data chunking, embeddings, vector store, Azure AI Search index) and fine-tuning.
Exam format, languages and policies when it was offered
Here is what Microsoft states:
Item | Detail |
|---|---|
Status | Retired June 1, 2026, 11:59 PM Central Standard Time |
Passing score | 700 or greater |
Delivery | Proctored |
Question styles | May include interactive components |
Languages | English, Japanese, Chinese (Simplified), Korean, German, Chinese (Traditional), French, Spanish, Portuguese (Brazil), Italian |
First retake | 24 hours after the first attempt |
Later retakes | Wait time varies; see Microsoft's retake policy |
Extra time | If the exam isn't offered in your language, you can request an extra 30 minutes |
For exam duration and session details, Microsoft points to its exam duration and experience page.
Microsoft also notes that most questions covered generally available features. Preview features could appear if they were commonly used.
Our voucher listing and its terms
Per Microsoft, the exam is retired. Do not buy our DP-100 voucher to sit DP-100. The terms below are our store policy for any voucher order.
Validity: 12 months.
Redemption: through Pearson VUE.
Delivery: the code is emailed within 12 hours of payment and also shown in your account.
Before delivery: you can cancel an order for a full refund.
After delivery: the code is non-refundable unless Pearson VUE rejects it as invalid, already used or expired on first use.
If a code is rejected, contact support within 14 days. Send your order number and a screenshot, and we replace or refund the code. Full terms are on our refund policy and delivery policy pages.
A six-week Azure ML skills plan at about 8 hours a week
This is a learning plan, not exam prep. Microsoft's certification page says no training is available for this exam. You can lean on the study guide's skill list, the free practice assessment and your own Azure subscription. The plan is our suggestion, not Microsoft's.
Week | Focus | Hours |
|---|---|---|
1 | Workspace, datastores, compute targets, data assets, environments | 8 |
2 | Automated ML (tabular, vision, NLP) and notebooks with MLflow tracking | 8 |
3 | Hyperparameter tuning and training scripts as jobs | 8 |
4 | Pipelines, model registration, online and batch endpoints | 8–10 |
5 | Model catalog, prompt flow, RAG with Azure AI Search, fine-tuning | 8–10 |
6 | Practice assessment, weak areas, exam sandbox | 6–8 |
If you are new to Azure, add a week at the start. If you want a lighter data foundation first, our DP-900 exam guide is a gentler step.
A worked example: one model from job to endpoint
Building one small flow end to end covers a large part of the "Train and deploy" area. This sketch uses the Azure CLI ml extension (v2). Replace the resource group and workspace names with your own.
az extension add --name ml
az group create --name rg-dp100 --location eastus
az configure --defaults group=rg-dp100 workspace=dp100-lab location=eastus
az ml workspace create --name dp100-lab --resource-group rg-dp100
az ml compute create --name cpu-cluster --type amlcompute --size Standard_DS3_v2 --min-instances 0 --max-instances 2 --resource-group rg-dp100 --workspace-name dp100-lab
az ml job create --file train-job.yml --resource-group rg-dp100 --workspace-name dp100-lab
az ml online-endpoint create --file endpoint.yml --resource-group rg-dp100 --workspace-name dp100-lab
az ml online-deployment create --file deployment.yml --all-traffic --resource-group rg-dp100 --workspace-name dp100-labAfter the job runs, open the run in the studio. Check that your MLflow metrics were logged, then register the model. Delete the endpoint when you finish so it does not keep billing you.
Then repeat the exercise with a pipeline and a batch endpoint. Online and batch deployment work differently, so do both.
Common mistakes when learning these skills
Studying only classic ML. Language model topics carried a 25–30% weight. Skipping prompt flow, RAG and fine-tuning leaves a large gap.
Reading without building. Click through the studio and the CLI yourself.
Ignoring MLflow. It shows up in notebooks, jobs and model registration.
Trusting old materials. The skills were updated on April 11, 2025, and Microsoft says Foundry materials were being updated. Check dates on any course or video.
Using brain dumps. They often reflect outdated content. Use Microsoft's free practice assessment and exam sandbox instead.
Leaving resources running. Compute clusters and endpoints can drain a trial budget fast.
Renewal status and where to go next
Microsoft's study guide says associate certifications expire annually and can normally be renewed with a free online assessment on Microsoft Learn. For this credential, the certification page says the renewal assessment is retired. Check your Microsoft Learn profile for the status of any credential you already hold.
To keep building skills, browse our AI and machine learning certifications for current options.
Sources
Questions about this topic
Can I still take the DP-100 exam?
No. Microsoft's study guide says the exam was retired on June 1, 2026, at 11:59 PM Central Standard Time. The certification page says the certification and its renewal assessment are retired. Check your Microsoft Learn profile for any credential you already hold.
Is there a free way to test myself on these skills?
Microsoft's certification page offers a free practice assessment that shows the style, wording and difficulty of questions. It also lists an exam sandbox so you can try the question types in the real interface.
Do I need to know Python for this role?
Microsoft's certification page names Python, Azure Machine Learning and MLflow as the tools behind the role. Expect to read and write training scripts and work with notebooks. Practice running scripts as jobs, not just reading them.
What if the exam was not in my language?
Microsoft listed ten languages, and English was always updated first. Localized versions followed about eight weeks later. If your preferred language was not offered, you could request an extra 30 minutes.
How soon could I retake the exam after failing?
Microsoft's page says you could retake it 24 hours after the first attempt. For later retakes the wait varied, so Microsoft points to its exam retake policy.
Which topics carried the most weight?
Train and deploy models, and optimize language models, each carried 25–30%. Together they made up more than half of the exam. They are also the best areas for hands-on practice.
