Google Cloud logoPMLE

Professional Machine Learning Engineer practice exams

Original questions written from the published objectives and cited to official documentation, never recalled exam content. How we verify, why not dumps.

What is ready now

2full-length forms120verified questions6of 6 domains with a full set
Exam code
PMLE
Cost
$200USD
Questions
50-60
Duration
120minutes
Passing score
Not published by Google Cloud
Format
Multiple choice and multiple select

Fail and the retake is another $200. Practising until you are ready is the cheapest part of this.

What the exam tests

Exam domains and official weightings

From the official Google Cloud exam guides. Put your study time where the weight is.

The percentage is the exam weighting. The bar is our verified questions in that domain against the best covered one, so a short bar is where the bank is thin.

6domains
  • Architecting low-code AI solutions13%
    16 verified
  • Collaborating within and across teams to manage data and models16%
    19 verified
  • Scaling prototypes into ML models21%
    25 verified
  • Serving and scaling models20%
    24 verified
  • Automating and orchestrating ML pipelines18%
    22 verified
  • Monitoring AI solutions13%
    14 verified

Where to focus: The operational half is where people underprepare. Serving, pipelines and monitoring together are 51 percent of the exam, against 34 percent for building and scaling models. Google's own section weights sum to 101 rather than 100, so treat every figure as approximate.

Every ML Engineer objective, with practice questions mapped to each one

What the exam actually asks you to do

Multiple choice and multiple select, in Google's own words. No labs, no console access, and nothing to configure during the exam.

Item formats

  • Multiple choice
  • Multiple response

Nothing here needs a lab. Reading carefully and eliminating options is the whole skill. Google Cloud, Professional Machine Learning Engineer exam details

ML Engineer practice exams

Full-length forms, 50 questions each, apportioned to the published domain weightings so every one stands alone as a representative sample. The questions in a numbered form are fixed, so two attempts on the same form are comparable.

Sit any of them under exam conditions, with the clock running and nothing revealed until you submit, or in study mode, with the full explanation after every answer. You choose when you start. 5 questions from across every form are free, with no account and no card.

Between sittings

The same bank the numbered forms are assembled from.

  • Open now

    Quick Practice

    All 120 verified questions, untimed, with the explanation after each answer.

  • Open now

    Domain Drill

    Every domain on its own, for the area a score report says is weakest.

  • Fills as you go

    Review Missed

    Re-asks the questions you got wrong. Nothing to review until you miss something.

Open ML Engineer practice

Where ML Engineer fits

An honest study plan

  1. 1. Read the official objectives first

    Download the official objectives from Google Cloud and skim every line. The exam can only test what’s listed there, it’s the contract.

  2. 2. Weight your study toward Scaling prototypes into ML models and Serving and scaling models

    Together the top two domains are 41% of the exam. Practise them until your accuracy is consistently above 80%.

  3. 3. Drill weak domains, then sit a numbered practice exam

    Use Domain Drill on your weakest areas, then sit a full-length form under exam conditions: 50 questions in 120 minutes, scored against the published pass mark. Book the real exam when you are consistently clearing that, not before.

Official free resources

Study from the source. These are Google Cloud’s own materials:

Official ML Engineer exam page & objectives ↗

Keep reading

Every guide and cost breakdown, by vendor

Frequently asked questions

How much of this is Vertex AI?

Most of it, under a new name. Google renamed Vertex AI to Gemini Enterprise Agent Platform during 2026, and the guide now uses that name throughout. The mechanisms are unchanged, so older study material describes the same exam.

Do I need to be able to write code?

Yes. The guide states the ML Engineer has strong programming skills and experience with distributed data processing tools. Scaling prototypes into ML models is the largest section at 21 percent and assumes you can read training code, not just describe it.

Is this an ML theory exam?

No. Roughly half the exam is operational: serving and scaling at 20 percent, pipelines at 18 and monitoring at 13. Knowing which serving pattern fits a latency requirement matters more than deriving an algorithm.

How long is it valid?

Two years, one less than the foundational and associate certifications. Renewal means passing the full exam again.

Ready to practice for PMLE?

Five questions from across every practice exam, plus 10 questions per certification in bank practice, with no account. Full explanations on every single one.