The Databricks Machine Learning Associate exam tests whether you can build, track and serve a model on Databricks; the Machine Learning Professional exam tests whether you can scale, test, monitor and roll out models in production. Both cost US$200, neither has a prerequisite, and most candidates should take the Associate first.
Last updated November 2026.
| Detail | ML Associate | ML Professional |
|---|---|---|
| Scored questions | 48 | 59 |
| Time limit | 90 minutes | 120 minutes |
| Question types | Multiple choice or multiple selection | Multiple choice |
| Fee | US$200 | US$200 |
| Recommended experience | 6 months | 1 year |
| Prerequisite | None | None |
| Objectives in the guide | 48 in 4 sections | 47 in 3 sections |
| Published section weights | No | Yes: 44 / 44 / 12 |
| Validity | 2 years | 2 years |
From the official exam guides: the Associate guide dated 1 Mar 2025 and the Professional guide dated 30 Sep 2025, both linked from their pages on the Databricks certification site. Objective counts are our line-by-line counts of each guide.
The two guides share vocabulary (MLflow, feature tables, Unity Catalog, Model Serving) but ask different questions about it.
Databricks Machine Learning, Data Processing, Model Development and Model Deployment. AutoML, exploring and preparing data, Spark ML pipelines, tuning and evaluation, MLflow tracking and the registry, and the basics of batch, streaming and real-time serving.
Model Development, MLOps and Model Deployment. Distributed training and tuning, custom PyFunc models, online and on-demand features, testing across environments, bundles, automated retraining, drift monitoring and safe rollouts.
The same topic shows the gap. On model promotion, the Associate expects you to know a model is registered in Unity Catalog and referenced by an alias. The Professional expects you to choose between deploying code and deploying models, decide which tests gate each environment, and roll a new version out behind a traffic split.
Classical modelling knowledge (choosing an algorithm, handling imbalance, reading a metric) is assumed by the Professional rather than tested head on, which is why skipping the Associate material is risky even though skipping the Associate exam is allowed.
You have under a year of Databricks ML work, you have not owned a model after deployment, or you want a credential in about six weeks. Its material is the foundation the Professional assumes.
You already run production ML on Databricks: scheduled retraining, a registered model with aliases, an endpoint or batch scorer, some monitoring. The Associate would mostly confirm what you do daily.
Taking both in sequence costs US$400 in exam fees. Preparation is roughly 5 to 7 weeks for the Associate and 7 to 9 weeks for the Professional at part-time pace, and the second is shorter if taken soon after the first because the platform vocabulary carries over. Those time estimates are ours.
Most candidates should take the Associate first, because the Professional assumes its material. Go straight to the Professional only if you already run production ML on Databricks, with retraining, a registered model and monitoring in place.
No. Neither exam has a prerequisite, so you can book the Professional directly. Databricks recommends six months of hands-on experience for the Associate and one year for the Professional.
The Professional has 59 scored questions in 120 minutes against 48 in 90 minutes, and it tests production work. It adds distributed training, testing across environments, automated retraining, drift monitoring and deployment rollouts.
Yes. Each exam costs US$200 per attempt, so taking both in sequence costs US$400 in exam fees. Both certifications are valid for two years.
Short visual chapters and timed practice exams for each certification. The Professional course is free to open, and the Associate course starts with a free unit.
Open the ML Professional course →