- The standard exam is 50-60 multiple-choice and multiple-select questions in two hours, for a $200 USD fee plus applicable tax.
- Scaling prototypes into ML models (~21%) and Serving and scaling models (~20%) are the two heaviest sections.
- Pearson VUE replaced Kryterion; Pearson testing began March 2, 2026, via test center or OnVUE.
- Certification lasts two years, and renewal means retaking the standard exam within a 60-days-before to 30-days-after window.
The Exam at a Glance
This page condenses the Professional Machine Learning Engineer credential into facts you can scan in minutes. It covers the Google Cloud certification administered by Google Cloud, and nothing else that happens to share the abbreviation. If you are still orienting yourself, start with What Is PMLE Certification? and come back here for the review sheet.
| Item | What to Remember |
|---|---|
| Format | 50-60 multiple-choice and multiple-select questions |
| Time | Two hours; remote check-in is separate from the exam clock |
| Fee | $200 USD plus applicable tax; no member/non-member distinction |
| Languages | English and Japanese |
| Delivery | Pearson VUE test center or OnVUE online proctoring, accessed through CM Connect |
| Result type | Pass/fail across the whole exam, not separate domain passes |
| Validity | Two years |
| Objective source | Exam guide as of June 1, 2026 (five pages) |
Six Sections and Their Weights
The exam guide lists six official examination sections with approximate weights. These are published as approximate figures, and they add up to 101% by simple arithmetic, so they should not be read as exact question allocations. The sections describe what is tested; they are not a preparation curriculum. For a longer treatment, see PMLE Exam Domains 2026: Complete Guide to All 6 Content Areas.
| Section | Approx. Weight |
|---|---|
| 1. Architecting low-code AI solutions | ~13% |
| 2. Collaborating within and across teams to manage data and models | ~16% |
| 3. Scaling prototypes into ML models | ~21% |
| 4. Serving and scaling models | ~20% |
| 5. Automating and orchestrating ML pipelines | ~18% |
| 6. Monitoring AI solutions | ~13% |
Two takeaways from the table: Sections 3 and 4 together carry roughly two-fifths of the exam, and Sections 1 and 6 are the lightest at about 13% each. Light does not mean skippable, since the exam is scored as a single pass/fail result and a weak area can still cost you points you needed.
Domain-by-Domain Must-Know Topics
The current exam guide uses Gemini Enterprise Agent Platform terminology and reflects updated generative-AI security and evaluation content, including Model Armor and LLM-as-a-judge. If you studied from older material, check it against the June 2026 guide, because older headings (such as a separate framing-ML-problems section) are not how the objectives are organized now.
Domain 1: Architecting low-code AI solutions (~13%)
Know when a managed or low-code path is the right architectural choice and when it is not.
- Match business needs to pre-built, AutoML-style, or custom approaches
- Understand trade-offs among speed, control, cost and maintainability
- Recognize where generative-AI tooling fits in a low-code architecture
- Be able to justify a choice, since scenario questions reward reasoning over recall
Domain 2: Collaborating within and across teams to manage data and models (~16%)
This section is about the shared workflow around data and models, not just individual notebooks.
- Data preparation and exploration across teams
- Experiment tracking, model versioning and reproducibility
- Governance and handoffs between data science and engineering roles
- Reading short Python and SQL snippets to decide what a step is doing
Domain 3: Scaling prototypes into ML models (~21%)
The heaviest section. It tests whether you can turn a working prototype into something production-grade.
- Training at scale, including distributed training considerations
- Choosing training infrastructure and accelerators sensibly
- Feature handling, evaluation approaches and tuning
- Generative-AI evaluation concepts, including LLM-as-a-judge
Domain 4: Serving and scaling models (~20%)
The second-heaviest section is about getting predictions to users reliably.
- Batch versus online prediction and when each fits
- Endpoint scaling, latency and cost trade-offs
- Deployment strategies and safe rollout of new model versions
- Generative-AI serving concerns, including protective controls such as Model Armor
Domain 5: Automating and orchestrating ML pipelines (~18%)
Repeatability is the theme: training and deployment as pipelines, not manual steps.
- Pipeline components, orchestration and scheduling
- Continuous training and deployment concepts
- Metadata, lineage and triggers for retraining
- Testing and validation gates within pipelines
Domain 6: Monitoring AI solutions (~13%)
What happens after deployment: detecting that something has gone wrong.
- Skew and drift detection and what should trigger action
- Performance, reliability and cost monitoring
- Monitoring generative-AI behavior and output quality
- Responsible-AI concerns in production
Key Takeaway
The exam rewards end-to-end lifecycle thinking. Sections 3, 4, 5 and 6 together form a chain from training through serving to automation and monitoring, so study them as one story rather than four isolated topics.
Question Style and What the Format Rewards
All questions are multiple-choice or multiple-select. There is no direct coding assessment, so you will not be asked to write code in a live environment. However, Python and SQL proficiency still matters because some questions include code snippets you must interpret. Think of coding knowledge as reading comprehension for the exam rather than a hands-on task.
- Scenario-driven stems: Expect a business or technical situation followed by a decision. The best answer usually balances constraints such as cost, latency, maintainability and team skill.
- Multiple-select traps: Read the instruction for how many answers to choose. Partial reasoning that picks one right and one plausible-but-wrong option will not help.
- Managed-first bias: When two options both work, the one that reduces operational burden on Google Cloud is frequently the intended answer, unless the scenario states a requirement that rules it out.
- Sample material: Google publishes an official sample question form, but performance on samples does not predict your result and does not establish topic coverage. Use it to see the style, not to judge readiness.
To gauge how demanding this style feels, read How Hard Is the PMLE Exam? Complete Difficulty Guide 2026, and drill the format on our PMLE practice tests.
Registration, Fees and Delivery Mechanics
The provider changed in 2026. Kryterion's last testing date was February 22, 2026, and Pearson VUE testing began March 2, 2026. Candidates schedule through CM Connect and choose between a Pearson VUE test center and OnVUE remote proctoring. Cancel or reschedule at least 24 hours before your appointment.
Eligibility in one paragraph
You must be at least 18. There is no prerequisite degree, credential, mandatory training hours, references or professional-experience minimum. Google's guidance recommends three-plus years of industry experience including at least one year designing and managing Google Cloud solutions, though a separate preparation paragraph phrases it as three-plus years with Google Cloud. Both are recommendations, not admission requirements. Details live in PMLE Requirements 2026: Eligibility, Prerequisites & How to Qualify, and scheduling specifics are in PMLE Exam Dates 2026: Testing Windows, Deadlines & Scheduling.
Test-Day Rules You Cannot Improvise
The exam is closed book. No notes, paper, books, writing instruments, phones or AI assistance are allowed. A separate calculator entitlement and adaptive delivery are not verified, so do not plan around either.
- OnVUE identity check: Have valid identification ready and expect a verification step before the exam begins.
- Workspace: A private, clear workspace is required.
- System test: Run the system test ahead of exam day, not minutes before.
- Continuous camera presence: You must stay on camera, and there are no ordinary breaks during the two hours.
- Accommodations: If you need approved accommodations, arrange them separately and early.
Results, Retakes and Renewal
Results
You receive a provisional pass/fail result immediately, with official confirmation normally taking 7-10 days. Scoring is across the whole exam, not by domain, so you cannot pass individual sections. Candidates who are unsuccessful on the standard exam can obtain section feedback, though Google notes exclusions for beta and renewal exams. For what the score represents, see PMLE Passing Score 2026: Exactly What You Need to Pass; for outcome data, see PMLE Pass Rate 2026: What the Data Shows.
Retake policy
| After... | Minimum Wait |
|---|---|
| First unsuccessful attempt | 14 days |
| Second unsuccessful attempt | 60 days |
| Third unsuccessful attempt | 365 days |
The cap is four attempts in two years, regardless of language or delivery method. Each retake is paid, so a failed attempt is both a time and a money cost.
Renewal
- Validity is two years.
- Renewal uses the standard exam; there is no shorter renewal exam and no continuing-education route, and no fixed CPE/CEU-hour requirement is published.
- The renewal window opens 60 days before your inactive date and extends up to 30 days afterward, with no promise of active status during that post-expiration gap.
- A conditional 50%-off renewal code can bring the base fee to a calculated $100 before tax when you are eligible.
- A documentation wrinkle exists: Google's renewal help says renewal adds two years to the inactive date, while the certification terms use the passing date unless otherwise specified. Confirm your exact expiration date in your account rather than assuming.
A Weight-Driven Review Plan
The one place a schedule helps is allocating time by section weight. Front-load the heavy sections while your energy is highest, then use the final week for the lifecycle chain and timed practice. This is an illustration; adapt it to your own background, and see the PMLE Study Guide 2026: How to Pass on Your First Attempt for a fuller approach.
Scaling prototypes into ML models (~21%)
- Training at scale, tuning and evaluation, including generative-AI evaluation
- Write down every service decision rule you rely on
Serving and scaling models (~20%)
- Batch versus online serving, scaling and rollout patterns
- Model Armor and protective controls for generative workloads
Pipelines (~18%) and data/model collaboration (~16%)
- Orchestration, continuous training, metadata and lineage
- Versioning, reproducibility and Python/SQL snippet reading
Low-code (~13%), monitoring (~13%) and timed practice
- Managed-versus-custom decisions and drift/skew response
- Full timed sets on the practice test site, reviewing every miss
Who Hires for This Credential
The certification maps to ML engineering work on Google Cloud: teams that take models from experimentation to production, run pipelines, and monitor deployed systems. Roles that commonly reference it include machine learning engineer, MLOps engineer, and data scientist positions that carry production responsibility. Organizations that build on Google Cloud are the most natural fit, and consulting and partner firms often value certifications as evidence of platform skill. Market detail is covered in PMLE Jobs, compensation context in PMLE Salary Guide 2026: Complete Earnings Analysis, and the cost-benefit question in Is the PMLE Certification Worth It? Complete ROI Analysis 2026.
Quick Answers (FAQ)
The standard exam has 50-60 multiple-choice and multiple-select questions in two hours. The range is not a fixed count, and Google does not disclose how many questions are scored versus unscored.
Scaling prototypes into ML models is largest at approximately 21%, followed by Serving and scaling models at approximately 20%. The published weights are approximate and total 101%, so they should not be treated as exact question counts.
No. There is no direct coding assessment. Python and SQL proficiency helps you interpret code-snippet questions, but no hands-on coding task is required.
The fee is $200 USD plus applicable tax. Retakes are paid, limited to four attempts in two years, with 14, 60 and 365-day waits after successive failures.
It is valid for two years. Renewal uses the standard exam within a window that opens 60 days before the inactive date and runs up to 30 days after, and a conditional 50%-off code may apply. For background on the credential's meaning, see What Is PMLE?