GenAI vs ML Associate: Which Databricks Exam?

The short version

Take the Generative AI Engineer Associate if your work is LLM applications, especially retrieval-augmented generation (RAG). Take the Machine Learning Associate if your work is training and deploying classic models such as classifiers and regressors. They are separate exams with little overlap in the question topics. If you do both kinds of work, the order depends on your current project, and taking both is reasonable. Facts here come from the two Databricks exam guides and certification pages, checked October 22, 2026.

Side by side

Item	Generative AI Engineer Associate	Machine Learning Associate
Level	Associate	Associate
Audience	Engineers who build and deploy LLM applications	People who do basic machine learning tasks on Databricks
Scored questions	45	48
Time	90 minutes	90 minutes
Sections	6: Design, Data Preparation, Application Development, Assembling and Deploying, Governance, Evaluation and Monitoring	4: Databricks Machine Learning, data processing (ML Workflows), Model Development, Model Deployment
Biggest section	Application Development, 30%	Databricks Machine Learning, 38%
Passing score	Not published	Not published
Languages	English, Japanese, Brazilian Portuguese, Korean	The same four
Delivery	Exam guide: online proctored; page: online or test center	Online proctored or test center
Prerequisites	None required; training recommended	None required; training recommended
Recommended experience	About six months of hands-on work	About six months of hands-on work
Validity	2 years, renew with the current exam	2 years, renew with the current exam

Both guides give the same retake rule through the Databricks FAQ: 14 days between attempts.

What each exam covers

The Generative AI exam is about building a pipeline around a language model. You prepare documents, chunk them, put them in a Vector Search index, assemble a chain, serve it, protect it, and then evaluate and monitor it. See the overview and domains explained.

The Machine Learning exam is about the classic workflow. You explore data, prepare features, choose an algorithm, tune it, track runs with MLflow, register a model in Unity Catalog, and deploy it for batch, streaming or real-time use. See the Machine Learning Associate overview.

Where they overlap

Both use MLflow for tracking and registering models, and Unity Catalog for governance.
Both involve Model Serving for deployment.
Both expect Python, with SQL appearing for non-ML code.
Both ask scenario questions where you choose a design that fits a constraint.

The overlap is in the platform, not in the question content. Passing one does not prepare you for the other beyond platform familiarity.

Where they differ

Kind of model. The ML exam deals with trained models and metrics such as classification and regression scores, tuning and cross-validation. The Generative AI exam deals with prompts, embeddings, retrieval and judges.
Data work. The ML exam covers cleaning, imputing and encoding features. The Generative AI exam covers text extraction, chunking and retrieval metrics.
Risk topics. The Generative AI exam has a Governance section on guardrails, masking and licensing. The ML exam has no equivalent section.
Newer features. The Generative AI outline names Agent Bricks, MCP servers and custom scorers. Because the guide is dated March 18, 2026 and the ML guide March 1, 2025, expect the Generative AI exam to reflect more recent platform changes.

How to choose

Ask three questions:

What do you build? LLM chat, search or agents: Generative AI. Prediction on tables: Machine Learning.
What does your employer or client ask for? Credentials are only worth the effort if someone values them. See is it worth it.
What can you already explain? Check the prerequisites self-check. If you cannot yet explain embeddings and chunking, the Generative AI exam will be a long climb.

A common path is the Machine Learning Associate first if you are new to data science on Databricks, because it builds the platform basics. That is our suggestion, not a Databricks requirement. Databricks lists neither exam as a prerequisite for the other.

Taking both

If you need both, study them in separate blocks. Do not blend their notes, because similar words have different meanings: “evaluation” in the ML exam means model metrics on test data, while in the Generative AI exam it can mean judges that grade responses. Space the exams out enough to book each one with its 14-day retake margin in mind.