Prepare for Google’s Professional Machine Learning Engineer exam by connecting data quality, model evaluation, serving and monitoring decisions. This guide follows the June 1, 2026 outline and shows how explicit precision and recall requirements can change which model you select.
Which exam does this guide cover?
This is the Google Cloud Professional Machine Learning Engineer credential, covering traditional and generative AI solutions from data and training through production and monitoring.
The applicable guide is effective June 1, 2026 and uses current Google Cloud product names, including Gemini Enterprise Agent Platform. Use its objective lists to check the coverage of older materials.
There is no prerequisite certification. Google recommends three or more years of industry experience, including one or more years designing and managing Google Cloud solutions. The general certification FAQ requires candidates to be at least 18. Use the objectives to assess your own preparation gaps.
Sources: Professional Machine Learning Engineer Certification exam guide (as of June 1, 2026) — Google Cloud exam guide PDF; Professional ML Engineer Certification | Learn | Google Cloud — Certification page; Google Cloud Certification FAQs - Cloud Certification Help — Official FAQ
What are the six current exam areas?
The published approximate weights are 13%, 16%, 21%, 20%, 18% and 13%. They total 101%; the table preserves them as printed.
Use the full guide for every sub-objective. For data practitioners, the study focus below connects those areas to observable inputs, decisions and results.
| Exam section | Approximate weight | Data-workflow study focus |
|---|---|---|
| 1. Architecting low-code AI solutions | ~13% | Compare BigQuery ML, AutoML, AI APIs and model options against a stated task |
| 2. Collaborating within and across teams to manage data and models | ~16% | Review exploration, preprocessing, features, privacy, notebooks, experiments and model lineage |
| 3. Scaling prototypes into ML models | ~21% | Explain training-data organization, model choice, tuning, custom training and hardware requirements |
| 4. Serving and scaling models | ~20% | Review batch/online inference, versions, rollout comparisons and serving requirements |
| 5. Automating and orchestrating ML pipelines | ~18% | Trace validation, preprocessing consistency, orchestration, retraining and automation |
| 6. Monitoring AI solutions | ~13% | Interpret evaluation evidence, skew and drift; review explainability, responsible AI and security |
Practice comparing models against a requirement
Precision is TP / (TP + FP), recall is TP / (TP + FN), and accuracy is (TP + TN) / total.
TP means true positives, FP false positives, FN false negatives and TN true negatives. Define the positive class and evaluation set before using these formulas. The acceptance rule below is fictional and supplied solely for the exercise.
Practice exercise
Fraud is the positive class in a fixed set of 200 labelled transactions: 60 fraudulent and 140 legitimate. At the tested thresholds, model A produces TP40, FP10, FN20 and TN130; model B produces TP50, FP25, FN10 and TN115. The same 200 transactions are evaluated by both models. The only acceptance criteria are recall of at least 80% and precision of at least 65%. Which model meets both, and why would choosing the highest accuracy give the wrong answer?
Show answer
Only B meets both criteria. A has precision 80%, recall approximately 66.7% and accuracy 85%. B has precision approximately 66.7%, recall approximately 83.3% and accuracy 82.5%.
A’s recall is 40/60, below 80%. B’s recall is 50/60 and its precision is 50/75, so both meet the stated limits. A has more correct predictions overall: 170/200 versus B’s 165/200. That higher accuracy does not satisfy the missing recall requirement. This establishes B’s acceptance only under the fictional rule and supplied outputs; it does not establish global superiority, calibration quality or production approval.
Build a study plan around the data workflow
Use the current guide, Google’s Machine Learning Engineer learning path, official samples and technical documentation to choose practice tasks.
Samples show format and example content; they do not cover the whole syllabus or predict your result. Use an authorized learning environment for hands-on tasks, checking access and costs before creating resources.
- Data: identify what each field, label and missing value means before choosing preprocessing or evaluation steps.
- Modeling: compare the listed modeling approaches and training options against explicit requirements; verify service capabilities in current docs.
- Evaluation: calculate both models’ metrics before reading the answer. Separate measured results from acceptance criteria and conclusions that remain untested.
- Serving and pipelines: trace how data transformations, model versions, validation and rollout decisions connect across a small learning example.
- Monitoring: review the guide’s skew, drift, explainability, security and generative-AI evaluation objectives and the evidence each requires.
- Before booking: revisit gaps, confirm the June 2026 guide remains current, and check appointment requirements.
Sources: Professional Machine Learning Engineer Certification exam guide (as of June 1, 2026) — Google Cloud exam guide PDF; Professional ML Engineer Certification | Learn | Google Cloud — Certification page; Classification: Accuracy, recall, precision, and related metrics | Machine Learning | Google for Developers — ML crash course
What should you expect on exam day?
The standard Professional Machine Learning Engineer exam lasts two hours and has 50–60 multiple-choice and multiple-select questions. Registration costs USD200, plus applicable tax. The exam is offered in English and Japanese.
Choose a testing center or Pearson OnVUE remote proctoring. Remote exams are closed book and normally permit no breaks. Check the current ID, workspace, system and appointment instructions; arrange any needed accommodation before scheduling.
Your initial pass/fail result is provisional. Google says to allow 7–10 days for official confirmation. It does not report a numerical score; unsuccessful standard-exam candidates receive section-level feedback to help guide further study.
Professional candidates may make up to four attempts within two years. After a first failure, wait at least 14 days; after a second, 60 days; after a third, 365 days. Attempts count across languages and delivery methods, and each registration is paid.
Sources: Professional ML Engineer Certification | Learn | Google Cloud — Certification page; Google Cloud Certification FAQs - Cloud Certification Help — Official FAQ; Retake Policy - Cloud Certification Help — Official policy page; Online Proctored Exams - Cloud Certification Help — Official online testing page
How does renewal work?
Professional certification is initially valid for two years. The exam renewal window opens 60 days before your inactive date, and exam renewal adds two years to that date.
The current renewal table does not offer this credential a shorter renewal exam or designated Google Skills continuing-education route. Use the standard exam and check your CM Connect profile for eligibility and dates.
Google permits renewal within 30 days after the inactive date, subject to the applicable exam options. More than 30 days later, recertification requires the standard exam at full list price.
Sources: Professional ML Engineer Certification | Learn | Google Cloud — Certification page; Google Cloud Certification FAQs - Cloud Certification Help — Official FAQ; Certification Renewal Options - Cloud Certification Help — Official renewal page
Sources
Facts checked against official Google Cloud pages:
- Professional Machine Learning Engineer Certification exam guide (as of June 1, 2026)
- Professional ML Engineer Certification | Learn | Google Cloud
- Google Cloud Certification FAQs - Cloud Certification Help
- Retake Policy - Cloud Certification Help
- Certification Renewal Options - Cloud Certification Help
- Online Proctored Exams - Cloud Certification Help
- Classification: Accuracy, recall, precision, and related metrics | Machine Learning | Google for Developers