TL;DR: Google published a new Professional Machine Learning Engineer exam guide dated June 1, 2026. The format did not change: 50 to 60 questions, 2 hours, $200 plus tax. The content did. Google’s certification page says the exam now reflects the move from Vertex AI to the Gemini Enterprise Agent Platform, an updated data and analytics stack, and a focus on Google Cloud native solutions. The guide lists six sections, led by “Scaling Prototypes Into ML Models” at about 21% and “Serving and Scaling Models” at about 20%. Be careful with study material: some popular summaries use different weights and older product names, and a few are out of date. This guide uses Google’s own documents first and labels everything else.
If you are preparing for this exam, or you bought a course a few months ago, the main risk is studying the wrong version. This post shows what Google says, where other sources disagree, and how to spend your study hours.
Quick Facts
| Item | What the sources say |
| Exam | Google Cloud Professional Machine Learning Engineer |
| Exam guide date | June 1, 2026 (Google’s exam guide PDF) |
| Questions | 50 to 60, multiple choice and multiple select |
| Duration | 2 hours |
| Fee | $200 plus applicable taxes |
| Languages | English and Japanese |
| Delivery | Online proctored or at a testing center, through Pearson VUE |
| Prerequisites | None required |
| Recommended experience | 3+ years in industry, including 1+ year designing and managing solutions on Google Cloud |
| Coding | Not assessed directly. Minimum proficiency in Python and SQL is enough to read code snippets |
| Sections | 6 (weights listed below) |
| Certification validity | 3 years for Professional certifications, per Google’s renewal FAQ. One third-party source says 2 years |
| Reported average pay | About $130,802 for Google Cloud Machine Learning Engineer roles (ZipRecruiter, as of October 5, 2026) |
What Google Says Changed
Google’s PMLE page has a “Recent Updates” note. In plain terms it lists three things:
- A transition from Vertex AI to the Gemini Enterprise Agent Platform
- Updates to Google Cloud’s data and analytics stack
- A focus on Google Cloud native solutions
The June 1, 2026 exam guide backs this up. Many of its “key products” are written as “Gemini Enterprise Agent Platform” or the shorter “Agent Platform,” for example, Agent Platform AutoML, Agent Platform Pipelines, and the Agent Platform Feature Store.
What Google does not say on those pages: which earlier exam guide this one replaced, what the old weights were, or when the previous version stopped being used. We could not find an official changelog. Anything that claims exact “old versus new” weights should be treated as a third-party estimate.
The Product Name Ledger
This is the most useful way to avoid wasting study time. Older books, videos, and practice sets use older names. The exam guide uses the new ones.
| Older name used in many study resources | Name used in Google’s June 2026 guide | Why it matters |
| Vertex AI | Gemini Enterprise Agent Platform (also written “Agent Platform”) | The exam guide and Google’s page use the new name. Search for it when you study |
| Vertex AI Pipelines | Agent Platform Pipelines | Pipeline questions are about 18% of the exam |
| Vertex AI Feature Store | Agent Platform Feature Store | Appears in the collaboration and serving sections |
| Vertex AI AutoML | Agent Platform AutoML | Appears in the low-code and scaling sections |
| Vertex AI Model Garden | Model Garden | Listed under low-code AI and serving |
A caution: the table above compares names as they appear in older community material against names in the new guide. Google’s page confirms the move from Vertex AI to the Gemini Enterprise Agent Platform. We did not find an official side-by-side rename list for every product, so do not assume every older feature kept the same behavior under a new name.
The Six Sections and How to Spend Your Hours
These are the sections in Google’s June 1, 2026 guide, as we read it. The percentages add up to about 101% because of rounding in the guide.
| # | Section | Weight | Key products named in the guide |
| 3 | Scaling Prototypes Into ML Models | ~21% | Agent Platform AutoML, BigQuery ML, Agent Platform Pipelines, Kubeflow, Cloud Storage |
| 4 | Serving and Scaling Models | ~20% | Agent Platform, Model Garden, Cloud Run, GKE, Feature Store, Model Registry |
| 5 | Automating and Orchestrating ML Pipelines | ~18% | Agent Platform Pipelines, Managed Service for Apache Airflow, Ray, Cloud Build |
| 2 | Collaborating Within and Across Teams | ~16% | Feature Store, Workbench, Colab Enterprise, Experiments, Pipelines, Kubeflow Pipelines, ML Metadata |
| 1 | Architecting Low-Code AI Solutions | ~13% | BigQuery ML, AutoML, Model Garden, Document AI API, Vision API, Translate API, Gemini, Imagen, Veo |
| 6 | Monitoring AI Solutions | ~13% | Model Monitoring, Model Armor |
A simple study order
Weight alone is not the best guide. Here is a way to combine weight with how much of the section is likely to be new to you.
| Priority | Section | Reason |
| 1 | Scaling Prototypes Into ML Models | Highest weight. Covers cost, complexity, latency and hardware choices |
| 2 | Serving and Scaling Models | Second highest. Batch versus online inference and scaling decisions |
| 3 | Automating and Orchestrating ML Pipelines | Third highest. Includes retraining automation and newer orchestration tools |
| 4 | Monitoring AI Solutions | Smallest weight, but Model Armor and AI risk topics are easy to skip if you studied older material |
| 5 | Low-Code AI Solutions | Includes foundation model products such as Gemini, Imagen and Veo |
| 6 | Collaborating Within and Across Teams | Notebooks, experiments and feature management you may already know |
Adjust this for your background. A data scientist who has never run production serving should flip sections 4 and 6. A platform engineer who has never trained a model should start with section 3.
Where Sources Disagree
This is where we need to be careful. Three types of source conflict showed up.
1. Section weights
| Source | Low-code | Collaboration | Scaling prototypes | Serving | Pipelines | Monitoring |
| Google exam guide (June 1, 2026) | ~13% | ~16% | ~21% | ~20% | ~18% | ~13% |
| Third-party summary A (pdfquiz) | ~13% | ~16% | ~21% | ~20% | ~18% | ~13% |
| Third-party summary B (certquests) | ~12% | ~16% | ~18% | ~20% | ~22% | ~12% |
Summary A matches Google’s guide. Summary B puts pipelines at the top with about 22% and uses different numbers for several sections. Use Google’s guide.
2. Platform naming
| Source | What it says |
| Google’s PMLE page and exam guide | Uses Gemini Enterprise Agent Platform |
| Third-party summary B (certquests) | Says all training and serving content is “Vertex AI-native” |
| A certification guide page (certalyze) | Lists Vertex AI and makes no mention of the rename. Its page shows a last review date of January 15, 2025, so it appears out of date |
The older Vertex AI wording is likely from before the update. Do not treat it as a competing official position.
3. Certification validity and renewal
| Source | Validity | Renewal options |
| Google’s renewal FAQ | Professional certifications remain valid for 3 years. Renewal window opens 60 days before the inactive date | Shorter renewal exam adds 2 years (for eligible professionals). Standard exam adds 3 years. Continuing education is described for Professional Cloud Architect and Professional Data Engineer |
| Third-party summary A (pdfquiz) | Says valid for 2 years | Lists a retake, a shorter renewal exam, or Google Skills courses |
The FAQ we read lists the continuing education path for the Architect and Data Engineer certifications only, so we cannot confirm that a courses-based renewal path applies to the PMLE. We also cannot explain the 2 year versus 3 year difference. Check Google’s renewal page for your own certification before you plan.
What the Exam Does and Does Not Test
| Topic | In scope | Notes |
| Choosing between BigQuery ML, AutoML and foundation models | Yes | Section 1 |
| Notebooks, experiment tracking, feature management | Yes | Section 2 |
| Training at scale, hardware choice, SDKs | Yes | Section 3 |
| Batch versus online inference, serving at scale | Yes | Section 4 |
| End to end pipelines and retraining automation | Yes | Section 5 |
| Risk identification, monitoring, troubleshooting | Yes | Section 6 |
| Writing code from scratch | No, per the guide | Python and SQL proficiency is for reading code snippets |
| Deep ML theory derivations | Not listed | Expect applied decisions, not proofs |
The guide’s own words about coding matter. If you have been worried about live coding, the guide says the exam does not directly assess it.
Generative AI on the Exam: What Is Confirmed and What Is Reported
This is the part many candidates ask about.
| Topic | Confirmed in Google’s guide as we read it | Reported only by third-party summaries |
| Gemini, Imagen, Veo | Yes, listed under low-code AI solutions | |
| Model Garden | Yes, listed under low-code AI and serving | |
| Model Armor | Yes, listed under monitoring | |
| Ray | Yes, listed under pipelines | |
| Prompt and context engineering | Reported as a new discipline in the exam | |
| Retrieval augmented generation | Reported | |
| Fine-tuning Gemini models in BigQuery ML | Reported | |
| Generative AI evaluation methods | Reported | |
| Responsible AI integrated into data management | Reported |
“Reported only” does not mean wrong. It means we did not find those exact topics named in the guide text we could read. The safe approach is to read the official guide yourself, then use third-party lists to find extra topics to practice.
Career and Salary
The exam is aimed at people who build and run ML systems on Google Cloud. The certification does not set your pay, but it signals practical cloud ML skills.
| Metric | Figure | Source |
| Average annual pay, Google Cloud Machine Learning Engineer | $130,802 | ZipRecruiter, as of October 5, 2026 |
| Middle 50% range | $111,500 to $149,000 | ZipRecruiter |
These are averages from one aggregator for a job title. They are not the pay of certified people specifically, and pay varies by city, seniority and employer. For wider comparisons, see our guides on artificial intelligence salary and jobs and the best cloud computing certifications.
If you are comparing ML credentials across vendors, these may help:
- DP-100 vs AI-300 for Microsoft’s data science and AI exam path
- AWS AIP-C01 exam guide for the AWS generative AI developer exam
- Databricks ML Associate vs ML Professional for a data platform angle
- Databricks certification path for the full set of Databricks exams
What It Costs
| Cost | Amount | Note |
| Exam fee | $200 plus applicable taxes | Google’s PMLE page |
| Renewal discount | Google’s renewal FAQ mentions a 50% off renewal discount code after initial certification | Check eligibility on Google’s page |
| Study material | Varies | Google’s learning path and an official Wiley study guide are linked from the PMLE page |
For a view across Google exams, see our GCP certification cost guide.
Common Mistakes
| Mistake | Why it hurts |
| Studying only Vertex AI material | The guide and Google’s page use the Gemini Enterprise Agent Platform name |
| Trusting a weights table that does not match Google’s guide | One summary places pipelines at the top, but Google’s guide puts scaling prototypes first |
| Assuming every old feature behaves the same under its new name | We found no official rename list |
| Skipping monitoring because it is the smallest section | Model Armor and AI risk topics are named in the guide, and the section is about 13% |
| Planning your renewal from a third-party summary | Validity and renewal paths differ between sources |
| Panicking about coding | Google’s guide says coding is not directly assessed |
What We Don’t Know Yet
| Open question | Why it matters |
| Which earlier exam version the June 1, 2026 guide replaced and its old weights | You cannot measure exactly how much shifted |
| Whether the 3 year or 2 year validity applies to the PMLE | Affects your renewal planning |
| Whether a courses-based renewal path applies to the PMLE | One third-party source says yes. The FAQ we read lists it for two other exams |
| Whether every older Vertex AI feature maps cleanly to a new name | Could affect practice questions written with old names |
| Whether Google will add or retire sections later | Exam guides can change |
How to Prepare, Step by Step
- Read Google’s exam guide first. It is dated June 1, 2026. Print the section list and mark what you know.
- Build a name map. Next to each older product name you know, write the new one used in the guide.
- Study by section weight. Start with scaling prototypes, then serving, then pipelines.
- Practice decisions, not definitions. Questions tend to ask which tool fits a cost, latency or scale constraint.
- Do a hands-on lab for each section. One lab per section is better than ten videos.
- Add generative AI topics from third-party lists, then verify them. Check anything unfamiliar against Google’s documentation.
- Take timed practice sets. Two hours for 50 to 60 questions is about two minutes each.
- Book your exam. Online proctoring or a testing center, through Pearson VUE.
FAQS
Did the Google Professional Machine Learning Engineer exam change in 2026?
Yes. Google’s page notes a recent update and the exam guide is dated June 1, 2026. The update reflects the move from Vertex AI to the Gemini Enterprise Agent Platform and changes to Google’s data and analytics stack.
Did the exam format change?
Not according to Google’s page. It is still 50 to 60 multiple choice and multiple select questions in 2 hours for $200 plus tax.
What are the six sections of the June 2026 guide?
Architecting low-code AI solutions (about 13%), collaborating within and across teams (about 16%), scaling prototypes into ML models (about 21%), serving and scaling models (about 20%), automating and orchestrating ML pipelines (about 18%), and monitoring AI solutions (about 13%).
Which section has the highest weight?
Scaling prototypes into ML models, at about 21% in Google’s guide. Some third-party summaries put pipelines first, but they do not match the official guide.
Is Vertex AI still on the exam?
Google’s page describes a transition from Vertex AI to the Gemini Enterprise Agent Platform, and the guide uses the new names. Older Vertex AI study material may still help with concepts, but you should learn the new names.
Does the exam test generative AI?
The guide lists Gemini, Imagen, Veo, Model Garden and Model Armor. Third-party summaries also report prompt engineering, retrieval augmented generation and generative AI evaluation. Confirm those topics against Google’s documentation.
Do I need to write code on the exam?
The guide says the exam does not directly assess coding. Minimum proficiency in Python and SQL is enough to interpret code snippets.
How much does the exam cost?
$200 plus applicable taxes, per Google’s PMLE page.
Are there prerequisites?
No. Google recommends 3+ years of industry experience, including 1+ year designing and managing solutions on Google Cloud.
How long is the certification valid?
Google’s renewal FAQ says Professional certifications remain valid for 3 years. One third-party source says 2 years, so check your own certification page.
Can I renew with courses instead of retaking the exam?
One third-party source says yes. The renewal FAQ we read describes a continuing education path for the Professional Cloud Architect and Professional Data Engineer, so we cannot confirm it for the PMLE.
How much do Google Cloud ML engineers earn?
ZipRecruiter lists an average of about $130,802 as of October 5, 2026, with a middle range of $111,500 to $149,000. This is a job title average, not a certification premium.
Should I wait for another update before studying?
We found no announcement of a further change. Study from the June 1, 2026 guide and recheck the guide the week before your exam.
