Microsoft DP-100: Designing and Implementing a Data Science Solution on Azure Practice Exam
DP-100 retired on June 1, 2026 and is succeeded by AI-300 (MLOps Engineer Associate). This practice set still covers all four DP-100 skill areas — from preparing an ML solution to training, deploying, and optimizing language models — useful reference for the Azure Machine Learning knowledge that carries into AI-300.
Start 24-hour free trial →DP-100 has retired — the successor is AI-300
Microsoft retired DP-100 on June 1, 2026, and the Azure Data Scientist Associate certification can no longer be earned or renewed through it. If you already hold the certification, it remains on your Microsoft transcript until its expiration date.
The successor is AI-300: Azure MLOps Engineer Associate, which reframes the role around operational machine learning — automation, CI/CD pipelines, monitoring, drift detection, and lifecycle governance — reflecting how production ML actually works today. Your DP-100 Azure Machine Learning knowledge transfers, but the MLOps and operational content is expanded. Confirm the current path on Microsoft Learn.
DP-100 exam at a glance
- Vendor
- Microsoft (Azure)
- Exam code
- DP-100 (retired June 1, 2026)
- Certification
- Microsoft Certified: Azure Data Scientist Associate
- Successor
- AI-300 (Azure MLOps Engineer Associate)
- Level
- Associate (role-based; data science / machine learning)
- Blueprint
- Final edition: skills measured as of April 11, 2025
- Format
- Typically 40–60 questions; multiple choice, multiple response, drag-and-drop, and case studies
- Duration
- About 100–120 minutes
- Passing score
- 700 of 1000 (scaled, not a raw percentage)
- Recommended experience
- Data science and machine learning on Azure using Azure Machine Learning, MLflow, Azure AI services (including Azure AI Search), and Azure AI Foundry; Python proficiency
- Cost (USD)
- Was $165 USD (associate tier) while active
- Validity
- One year; could be renewed free online while active — no longer renewable after retirement
Source: Microsoft Learn — DP-100 study guide (skills measured) and the Azure Data Scientist Associate certification page. DP-100 retired June 1, 2026; see AI-300 for the current path.
About the DP-100 Azure Data Scientist Associate certification
DP-100 validated that a data scientist could design and implement machine learning solutions on Azure — preparing an Azure Machine Learning workspace and compute, exploring data and running experiments with automated ML and notebooks, training and deploying models through pipelines and endpoints with MLflow, and (in its final edition) optimizing language models for AI applications using prompt engineering, retrieval-augmented generation, and fine-tuning. It earned the Microsoft Certified: Azure Data Scientist Associate certification with a single exam.
DP-100 retired on June 1, 2026 (see the notice above). Its successor, AI-300: Azure MLOps Engineer Associate, shifts the emphasis from designing and training models toward operationalizing them — MLOps automation, CI/CD, monitoring, and drift detection. This practice set remains a useful reference for the Azure Machine Learning fundamentals — workspaces, automated ML, pipelines, MLflow, and endpoints — that carry directly into AI-300. If you are new to the data track, the DP-900 Azure Data Fundamentals exam is the conceptual starting point, and AI-900 Azure AI Fundamentals covers the AI concepts.
Every practice question maps to one of the four DP-100 skill areas and links to the specific Microsoft Learn page it was derived from, so your weak spots become a focused reading list rather than a guess.
DP-100 skill areas and weights
Four skill areas in the final edition, with two tied as heaviest at 25–30% — training and deploying models, and the newer language-model optimization area. Microsoft published each area as a weighting range rather than an exact percentage; the ranges below are reproduced as Microsoft stated them.
Design an ML solution (dataset structure, compute specifications, development approach); create and manage an Azure Machine Learning workspace, datastores, compute targets, and Git integration; and create and manage assets — data assets, environments, and shared registries.
Use automated machine learning for tabular data, computer vision, and NLP; use notebooks for custom training (compute instances, data wrangling, feature store, MLflow tracking, model evaluation with responsible AI); and automate hyperparameter tuning (sampling methods, search space, primary metric, early termination).
Run model training scripts as jobs (consume data, configure compute and environments, track with MLflow, troubleshoot); implement training pipelines with custom components; manage models (MLmodel signature, feature specs, MLflow registration, responsible AI assessment); and deploy to online and batch endpoints.
The newest area, foreshadowing the AI/MLOps direction: prepare for optimization (deploy from the model catalog, benchmark, playground testing); optimize through prompt engineering and prompt flow; optimize through retrieval-augmented generation (chunking, embedding, vector and Azure AI Search index stores); and optimize through fine-tuning.
Source: Microsoft Learn — DP-100 study guide (final edition, skills measured as of April 11, 2025). Microsoft published weightings as ranges; training/deploying models and optimizing language models were tied as the heaviest areas.
Who DP-100 was for
The Azure Data Scientist Associate certification targeted practitioners building and operationalizing machine learning on Azure:
- Data scientists training, evaluating, and deploying models with Azure Machine Learning and MLflow.
- Machine learning engineers building training pipelines and managing online and batch endpoints in production.
- Applied AI practitioners optimizing language models through prompt engineering, RAG, and fine-tuning.
- Python developers moving into data science and MLOps roles on Azure.
DP-100 assumed real data-science experience — it was not an entry point. If you are new to the data track, the DP-900 Azure Data Fundamentals exam is the conceptual starting point, and AI-900 Azure AI Fundamentals covers the AI concepts. Because DP-100 has retired, plan toward AI-300 for the current credential. For the roles this skill set supports — with skills, tools, and salary ranges — see the AI Engineer career path.
What this DP-100 practice exam delivers
Learn mode
Get the correct answer, the reasoning, and a direct link to the exact Microsoft Learn page each question was derived from — immediately after each question. Built for the Azure Machine Learning workflow scenarios across workspaces, pipelines, deployment, and language-model optimization.
Exam mode
A timed run in the DP-100 shape — roughly 40 to 60 questions in about 100 to 120 minutes, including case studies — useful pacing practice for AI-300, which keeps a similar format.
Source-linked explanations
Every answer cites the exact Microsoft Learn documentation page it was built from — so you can verify Azure Machine Learning behavior, much of which carries directly into AI-300.
Score by skill area
Results break down by the four DP-100 areas — design and prepare, explore and experiment, train and deploy, and optimize language models — so practice tells you exactly which area to revisit.
Sample DP-100 practice questions
Ten free questions across the four DP-100 skill areas, with full explanations and source links to the Microsoft Learn pages each is derived from. The complete bank is available with the 24-hour trial.
A data science team needs a central Azure resource to organize datastores, compute, experiments, models, and endpoints for their machine learning work.
Which Azure resource should they create first?
- An Azure Machine Learning workspace
- An Azure Storage account only
- An Azure Kubernetes Service cluster
- An Azure Key Vault only
Show answer & explanation
Correct: A — an Azure Machine Learning workspace. The workspace is the top-level Azure ML resource that organizes datastores, compute targets, experiments, models, and endpoints — exactly the central container described. It is the first thing you create for Azure ML work.
Why not the others: a storage account (B) and Key Vault (D) are provisioned as part of a workspace but are not the ML container themselves; an AKS cluster (C) can serve deployments but is not where ML assets are organized.
Source: Microsoft Learn — Azure Machine Learning workspace → Further reading: PowerKram — DP-900 Azure Data Fundamentals →A team wants a reusable, versioned reference to a specific dataset so experiments across the workspace all point to the same governed data.
Which Azure Machine Learning capability should they use?
- A compute instance
- A data asset
- An online endpoint
- A hyperparameter sweep
Show answer & explanation
Correct: B — a data asset. Data assets provide reusable, versioned references to data in Azure ML, so experiments consistently point to the same governed dataset — exactly the requirement.
Why not the others: a compute instance (A) is a dev workstation; an online endpoint (C) serves deployed models; a hyperparameter sweep (D) tunes training. None is a versioned data reference.
Source: Microsoft Learn — What is Azure Machine Learning? →A data scientist wants Azure Machine Learning to automatically try many algorithms and preprocessing options to find the best-performing model for a tabular classification problem, with minimal manual coding.
Which capability should they use?
- Automated machine learning (AutoML)
- A batch endpoint
- MLflow model registration
- A managed online endpoint
Show answer & explanation
Correct: A — automated machine learning (AutoML). AutoML automatically trials algorithms and preprocessing to find the best model for tabular, vision, or NLP tasks with minimal manual coding — exactly the requirement.
Why not the others: a batch endpoint (B) and managed online endpoint (D) serve predictions from an already-trained model; MLflow registration (C) versions a model. None searches for the best model.
Source: Microsoft Learn — Automated machine learning → Further reading: PowerKram — AI-900 Azure AI Fundamentals →A team wants to track parameters, metrics, and artifacts across many training runs in a standard, open-source-compatible way inside Azure Machine Learning.
Which tool should they use for experiment tracking?
- Azure Key Vault
- Azure DNS
- MLflow
- Azure Bastion
Show answer & explanation
Correct: C — MLflow. Azure Machine Learning uses MLflow — the open-source standard — to track parameters, metrics, and artifacts across training runs, and to package and register models — exactly the tracking requirement.
Why not the others: Key Vault (A) stores secrets; Azure DNS (B) resolves names; Azure Bastion (D) provides secure VM access. None is an experiment-tracking framework.
Source: Microsoft Learn — MLflow and Azure Machine Learning →A model’s performance depends heavily on settings like learning rate and tree depth. The team wants Azure ML to systematically try combinations and stop underperforming runs early.
Which technique should they configure?
- A data asset
- A datastore
- A hyperparameter tuning (sweep) job
- A compute instance
Show answer & explanation
Correct: C — a hyperparameter tuning (sweep) job. A sweep job searches a defined hyperparameter space using a sampling method and a primary metric, with early-termination policies to stop poor runs — exactly the systematic tuning described.
Why not the others: a data asset (A) references data; a datastore (B) connects to storage; a compute instance (D) is a dev workstation. None tunes hyperparameters.
Source: Microsoft Learn — Hyperparameter tuning →A trained model must serve low-latency, real-time predictions to a web application, one request at a time, over a REST endpoint.
Which deployment target should you use?
- A batch endpoint
- A managed online endpoint
- A datastore
- An automated ML job
Show answer & explanation
Correct: B — a managed online endpoint. Managed online endpoints serve real-time, low-latency predictions over REST for request-at-a-time inference — exactly the requirement.
Why not the others: a batch endpoint (A) scores large datasets asynchronously, not real-time; a datastore (C) connects to storage; an AutoML job (D) trains models. None serves low-latency real-time inference.
Source: Microsoft Learn — Endpoints for inference → Further reading: PowerKram — AI-900 Azure AI Fundamentals →A team must score millions of records overnight on a schedule, where throughput matters far more than per-request latency.
Which deployment approach fits best?
- A managed online endpoint
- A real-time inference cluster
- A batch endpoint
- A compute instance
Show answer & explanation
Correct: C — a batch endpoint. Batch endpoints run asynchronous scoring jobs over large datasets where throughput matters more than latency — exactly the overnight bulk-scoring scenario described.
Why not the others: a managed online endpoint (A) and real-time cluster (B) optimize for low-latency single requests; a compute instance (D) is a dev workstation. None is the bulk asynchronous option.
Source: Microsoft Learn — Batch endpoints →A team wants a repeatable, multi-step workflow — data prep, training, evaluation, registration — that can be scheduled and reused, with data passed between steps.
Which Azure Machine Learning construct should they build?
- A single notebook cell
- A training pipeline with components
- A Key Vault secret
- A data asset
Show answer & explanation
Correct: B — a training pipeline with components. Azure ML pipelines chain reusable components into a repeatable, schedulable multi-step workflow, passing data between steps — exactly the requirement for prep, training, evaluation, and registration.
Why not the others: a single notebook cell (A) is not a reusable orchestrated workflow; a Key Vault secret (C) stores credentials; a data asset (D) references data. None orchestrates multi-step training.
Source: Microsoft Learn — Azure Machine Learning pipelines → Further reading: PowerKram — DP-900 Azure Data Fundamentals →A team wants a deployed language model to answer using the company’s own knowledge base, retrieving relevant chunks and supplying them as context rather than retraining the model.
Which optimization approach should they use?
- Increasing the model temperature
- Reducing the token limit
- Hyperparameter sweeping
- Retrieval-augmented generation (RAG)
Show answer & explanation
Correct: D — retrieval-augmented generation (RAG). RAG grounds a language model in your own data by retrieving relevant chunks (often via a vector or Azure AI Search index) and supplying them as context — exactly the requirement, and a core DP-100 language-model optimization method.
Why not the others: raising temperature (A) increases randomness; reducing the token limit (B) truncates output; hyperparameter sweeping (C) tunes training. None grounds a model in private knowledge.
Source: Microsoft Learn — Retrieval-augmented generation → Further reading: PowerKram — AI-900 Azure AI Fundamentals →A general-purpose language model must consistently produce output in a specialized domain style and format. Prompting alone is not reliable enough, and the team has a labeled dataset of ideal examples.
Which optimization approach is most appropriate?
- Selecting a larger context window
- Adding more few-shot examples only
- Switching to batch inference
- Fine-tuning the model
Show answer & explanation
Correct: D — fine-tuning the model. When prompting is not reliable enough and a labeled dataset of ideal examples exists, fine-tuning adapts the model’s weights to the desired domain style and format — exactly the scenario DP-100 tests under language-model optimization.
Why not the others: a larger context window (A) and more few-shot examples (B) are prompt-level tactics that the scenario says are insufficient; batch inference (C) is a deployment mode, not an optimization of behavior.
Source: Microsoft Learn — Model catalog and fine-tuning →Keep going: study & career resources
DP-100 has retired, but the Azure Machine Learning knowledge behind it carries directly into AI-300. Below it, the data-fundamentals exam builds the base — and the AI-engineer career track shows where these skills lead.
Deep dive: DP-100 retirement, the AI-300 successor, the four areas, and study value
Retirement and the AI-300 successor
DP-100 retired on June 1, 2026, and the Azure Data Scientist Associate certification can no longer be earned or renewed through it. Its successor is AI-300: Azure MLOps Engineer Associate, which reframes the role around operational machine learning — automation, CI/CD pipelines, monitoring, drift detection, and lifecycle governance. If you already hold the DP-100 certification, it stays on your transcript until it expires, but you cannot renew it. See the DP-900 Data Fundamentals path →
What changed: from building models to operationalizing them
DP-100 centered on designing, training, and deploying models with Azure Machine Learning. AI-300 shifts the emphasis toward MLOps: getting models reliably into production and keeping them healthy through automation, monitoring, and governance. DP-100 already pointed this direction by adding the language-model optimization area (model catalog, prompt flow, RAG, fine-tuning) to its final blueprint. Career Hub — AI Engineer →
The four skill areas
The final DP-100 blueprint covered designing and preparing an ML solution (workspaces, compute, assets), exploring data and running experiments (automated ML, notebooks, hyperparameter tuning), training and deploying models (jobs, pipelines, model management, online and batch endpoints), and optimizing language models (model catalog, prompt flow, RAG, fine-tuning). The last two areas were tied as the heaviest at 25–30% each. Much of this — Azure ML, MLflow, pipelines, endpoints, and RAG — maps directly onto AI-300. See the AI-900 Fundamentals path →
Why this practice set is still worth using
The exam is retired, but the underlying Azure Machine Learning knowledge is not. If you are preparing for AI-300, working through DP-100-style scenarios on workspaces, automated ML, pipelines, MLflow, endpoints, and RAG builds the foundation the new exam assumes — you then layer the deeper MLOps automation, monitoring, and governance content on top. Treat this set as an Azure ML fundamentals warm-up, not a substitute for AI-300-specific study. See the AZ-104 Administrator path →
Format and scoring (as it was)
DP-100 delivered roughly 40 to 60 questions in about 100 to 120 minutes, scored 700 out of 1000 to pass, with multiple choice, multiple response, drag-and-drop, and case studies, and historically an interactive lab or portal component. AI-300 keeps a similar associate-level shape, so pacing practice transfers even as the content emphasis moves toward operations. Azure Data Scientist Associate certification page →
Career outlook
Azure data-science and machine-learning skills remain in strong demand — the credential simply moved to AI-300 with an operational emphasis. Roles include data scientist, machine learning engineer, and MLOps engineer. Employers increasingly expect production ML skills: pipelines, deployment, monitoring, and lifecycle governance, which is exactly where the certification path has moved. Career Hub — AI Engineer →
Microsoft DP-100 (Azure Data Scientist Associate) exam FAQ
Is DP-100 retired, and what replaces it?
What were the DP-100 skill areas and weights?
Should I still study DP-100 material?
What was the DP-100 passing score and format?
What experience did DP-100 assume?
Start your free 24-hour DP-100 practice trial
Full access to the question bank, both study modes, source-linked explanations, and score-by-skill-area — a solid Azure Machine Learning foundation as you move toward AI-300. No credit card required.
Start free trial →