Databricks Certified Machine Learning Professional: The Complete 2026 Guide
Databricks ML Professional · Guide

Databricks Certified Machine Learning Professional: The Complete 2026 Guide

The Databricks Certified Machine Learning Professional exam tests whether you can run machine learning in production on Databricks. It has 59 scored multiple-choice questions in 120 minutes, costs US$200, and covers three weighted sections: Model Development (44%), MLOps (44%) and Model Deployment (12%), across 47 objectives.

Last updated November 2026.

The exam at a glance

DetailML Professional
Scored questions59, multiple choice
Time limit120 minutes
Registration feeUS$200
DeliveryOnline proctored or test center, English only
Test aidesNone allowed
PrerequisitesNone required; one year of hands-on experience recommended
Passing scoreNot published
Validity2 years

Facts are from the Databricks Machine Learning Professional certification page and the exam guide PDF linked from it (30 Sep 2025 edition). Confirm them there before you book.

The exam guide names no programming language, but its objectives and sample questions are written in Python and Spark ML, so expect to read PySpark, MLflow and Feature Engineering client code.

Who the exam is for

The Machine Learning Associate exam asks whether you can train, track and serve a model on Databricks. The Professional exam assumes that and asks what happens next: the data outgrows one node, the model has to be retrained without a person in the loop, a change has to be tested before it reaches production, and a drifting feature has to raise an alert before the business notices.

That makes it a fit for ML engineers and data scientists who already ship models and want to prove production judgment. There is no prerequisite, so you can book it without holding the Associate. The Associate vs Professional comparison covers which to take first.

The three sections and their weights

Unlike several Databricks guides, this one publishes official weights. Two sections carry 88% of the exam between them.

SectionWeightObjectives and questions
1. Model Development44%22 objectives, about 26 questions
2. MLOps44%20 objectives, about 26 questions
3. Model Deployment12%5 objectives, about 7 questions

Weights are official. Objective counts are a line-by-line count of the guide's bullets, and question counts are the weight applied to 59 scored questions, our arithmetic and not a published figure.

Section 1: Model Development (44%)

Four sub-headings. Spark ML: when distributed training is justified, pipelines, feature transformers, tuning with CrossValidator or TrainValidationSplit, evaluators, and batch or streaming scoring. Scaling and tuning: pandas UDFs and the Pandas Function API for one model per group, distributed Optuna logged to MLflow, Ray Tune, and the vertical vs horizontal and data vs model parallelism trade-offs. Advanced MLflow: nested runs, programmatic logging and custom PyFunc models. Advanced Feature Store: point-in-time lookups, scheduled feature pipelines, Online Feature Stores, streaming features and on-demand features.

Section 2: MLOps (44%)

Half of this section's 20 objectives sit under one sub-heading: drift detection and monitoring. It covers drift tests for numerical and categorical features, profile types, building a monitor, baseline vs consecutive drift, alerts, inference tables, custom metrics, slicing and endpoint health. The other half covers the model lifecycle (see deploy code vs deploy models), unit and integration testing, environment design, Declarative Automation Bundles and automated retraining.

Section 3: Model Deployment (12%)

Five objectives: blue-green vs canary rollouts, traffic splitting across served entities on Model Serving, registering custom PyFunc models with their artifacts, and deploying and querying endpoints with the MLflow Deployments SDK or the REST API. It is the smallest section but the densest per objective: roughly seven questions over five objectives.

Product names have moved since the guide was written

The guide tells candidates to study current, non-deprecated features, and several names it uses have since changed in the Databricks documentation. Learn the current name and recognise the older one:

How to prepare

Most candidates with Associate-level knowledge need 7 to 9 weeks; the study plan breaks that down week by week. Databricks publishes no passing score, so we use 70% on timed practice as our own target, and 85% on full mocks as the readiness bar. Both numbers are our guidance, not an official threshold.

FAQ

How many questions are on the Databricks Machine Learning Professional exam?

The exam has 59 scored multiple-choice questions and a 120-minute time limit. Databricks may add a small number of unscored questions that are not identified, with the extra time already included.

How much does the Databricks Machine Learning Professional exam cost?

The registration fee is US$200 per attempt. The certification is valid for two years, and recertifying means passing the current version of the exam again.

Do I need the Machine Learning Associate before the Professional?

No. The Machine Learning Professional exam has no prerequisites, so you can book it directly. Databricks recommends at least one year of hands-on experience with the machine learning tasks in the exam guide.

What is the passing score for the Databricks Machine Learning Professional exam?

Databricks does not publish a passing score for this exam. A practical target is 70% or more on timed practice exams and 85% on full-length mocks, which is our guidance and not an official threshold.

What topics does the Databricks Machine Learning Professional exam cover?

It covers three sections: Model Development at 44%, MLOps at 44% and Model Deployment at 12%. In practice that means Spark ML, distributed training and tuning, advanced MLflow and Feature Store, testing and environments, drift monitoring, and Model Serving rollouts.

Study it chapter by chapter

Fifty-two short visual chapters mapped to all 47 objectives, six unit exams and three full-length mocks. Every unit is free to open, and it works on a phone.

Open the ML Professional course →