The Google Cloud Professional Machine Learning Engineer exam is a two-hour test with 50 to 60 multiple-choice and multiple-select questions, and it costs $200 plus tax where applicable. Google updated it for the move from Vertex AI to Gemini Enterprise Agent Platform, so the biggest risk now is studying from material that uses the old names and an older blueprint.
This guide covers the current exam facts, what changed in the updated exam guide, a name translation table, a 6-week study plan weighted to the six sections, and a prompt that turns old study notes into current ones.
The exam at a glance
From Google’s official exam page and certification help center, as of October 2026:
| Detail | What Google says |
|---|---|
| Length | Two hours |
| Format | 50 to 60 multiple-choice and multiple-select questions |
| Registration fee | $200, plus tax where applicable |
| Languages | English, Japanese |
| Delivery | Online proctored from home or office, or at a test center (delivered through Pearson) |
| Prerequisites | None |
| Recommended experience | 3+ years of industry experience, including 1+ year designing and managing solutions on Google Cloud |
| Coding | Not directly assessed; minimum Python and SQL proficiency lets you read code snippets |
| Results | Pass or fail only, no numeric score. Provisional result on screen, confirmed within 7 to 10 days. |
| Validity | 2 years for Professional certifications |
What changed in the updated exam
Google’s exam page says the exam “was updated to reflect the transition from Vertex AI to Gemini Enterprise Agent Platform, updates to Google Cloud’s data and analytics stack, and prioritizes Google Cloud native solutions.” The current exam guide is dated June 1, 2026. Compared with the older, Vertex AI-era guide, here’s what stands out:
- New names everywhere. Vertex AI Pipelines, Workbench, Feature Store, Model Registry, Experiments and Model Monitoring now appear under the Gemini Enterprise Agent Platform (or “Agent Platform”) name. Cloud Composer appears as Managed Service for Apache Airflow.
- More generative AI. The guide now lists fine-tuning Gemini models using BigQuery, building and tuning solutions with Gemini, Imagen and Veo, optimizing Gemini-based apps for cost, latency and availability, and evaluating gen AI with LLM-as-a-judge.
- AI security by name. Section 6 now calls out data exfiltration, malicious prompting and sharing sensitive data with LLMs, and names tools such as regex, safety filters and Model Armor.
- Broader monitoring and rollout topics. Data drift and concept drift join training-serving skew and feature attribution drift, and canary deployments sit next to A/B testing. Continuous training joins CI/CD, and Ray on Agent Platform appears as an orchestration option.
- Some older topics are gone from the outline. TFX system design, hosting MLflow, hybrid or multicloud strategies, and Jenkins no longer appear in the current guide.
- Section weights shifted. Pipelines and automation dropped from about 22% in the older guide to about 18%, while scaling prototypes rose from about 18% to 21% and data and model collaboration from about 14% to 16%.
Old name, new name: a translation table
If your course, book or flashcards say “Vertex AI,” you can still use them. Just translate as you go. These pairs come from Google’s official name-changes page, except Cloud Composer, which Google’s docs now call Managed Service for Apache Airflow:
| Name in older material | Name in the current exam guide and docs |
|---|---|
| Vertex AI Platform | Gemini Enterprise Agent Platform |
| Vertex AI Workbench | Gemini Enterprise Agent Platform Workbench |
| Vertex AI Pipelines | Gemini Enterprise Agent Platform Pipelines |
| Vertex AI Feature Store | Feature Store on Gemini Enterprise Agent Platform |
| Vertex AI Model Registry | Model Registry on Gemini Enterprise Agent Platform |
| Vertex AI Experiments | Experiments on Agent Platform |
| Vertex AI Model Monitoring | Model Monitoring on Gemini Enterprise Agent Platform |
| Vertex AI Model Garden | Gemini Enterprise Agent Platform Model Garden |
| Vertex AI AutoML | AutoML on Gemini Enterprise Agent Platform |
| Vertex AI Endpoints | Gemini Enterprise Agent Platform Endpoints |
| Vertex AI Inference | Gemini Enterprise Agent Platform Inference |
| Cloud Composer | Managed Service for Apache Airflow |
The services didn’t disappear; the names changed. The official study guide book that Google links from the exam page still describes the Vertex AI platform, so the translation matters there too.
The six sections and what to study
| Section | Weight | Focus on |
|---|---|---|
| 1. Architecting low-code AI solutions | ~13% | BigQuery ML and AutoML model choices, AI APIs such as Document AI, Vision and Translate, picking models from Model Garden, tuning Gemini, Imagen and Veo, optimizing Gemini apps for cost and latency |
| 2. Collaborating within and across teams to manage data and models | ~16% | Choosing preprocessing tools (BigQuery, Dataflow, Spark, in-memory Python), Feature Store, handling PII, Workbench and Colab Enterprise, Experiments, ML Metadata, evaluating predictive and gen AI models |
| 3. Scaling prototypes into ML models | ~21% | Choosing model type and product, custom training vs AutoML vs Kubeflow on GKE, hyperparameter tuning, troubleshooting training, fine-tuning foundation models, CPU vs GPU vs TPU, data and model parallelism |
| 4. Serving and scaling models | ~20% | Batch vs online inference on Agent Platform, Cloud Run and GKE, prebuilt and custom containers, Model Registry, A/B and canary rollouts, public vs private endpoints, scaling the serving backend |
| 5. Automating and orchestrating ML pipelines | ~18% | Data and model validation, Agent Platform Pipelines, Managed Service for Apache Airflow, Ray, consistent preprocessing between training and serving, retraining policies, CI/CD/CT with Cloud Build |
| 6. Monitoring AI solutions | ~13% | Model Armor and safety filters, responsible AI and bias, explainability, Model Monitoring, skew and drift types, testing and evaluating gen AI solutions |
Because Google says the exam doesn’t directly assess coding, expect to be tested on choosing the right product or approach for a requirement. Learn when to use each service and why the alternatives are worse.
A 6-week study plan
This assumes you already work with ML and can give it 6 to 8 hours a week. Google recommends 3+ years of industry experience, so if you’re newer, double the timeline and spend more of it hands-on.
Week 1: Blueprint and low-code (Section 1)
- Read the exam guide end to end and mark every product you haven’t used.
- Train a BigQuery ML model and an AutoML model on the same small dataset, then compare when you’d choose each.
- Try one pre-trained API (Document AI or Vision) and compare two models in Model Garden for the same task.
Week 2: Data and collaboration (Section 2)
- Practice choosing between BigQuery, Dataflow and Spark for preprocessing at different scales.
- Set up a Workbench or Colab Enterprise notebook, log runs to Experiments, and create features in Feature Store.
- Review how to find and protect PII before it reaches training data or an LLM.
Week 3: Training at scale (Section 3)
- Run a custom training job and a hyperparameter tuning job.
- Learn when fine-tuning a foundation model beats prompting, and when a simpler model (ARIMA, a DNN) is the right call.
- Know CPU vs GPU vs TPU trade-offs and data vs model parallelism.
Week 4: Serving (Section 4)
- Deploy one model for online inference and run one batch inference job.
- Register versions in Model Registry and plan an A/B test and a canary rollout.
- Compare Agent Platform endpoints, Cloud Run and GKE for serving.
Week 5: Pipelines and monitoring (Sections 5 and 6)
- Build a small pipeline with validation, training and deployment steps, and trigger it from Cloud Build.
- Set up Model Monitoring and learn to tell skew, data drift, concept drift and attribution drift apart.
- Read how Model Armor screens prompts and responses for prompt injection, jailbreaks and sensitive data.
Week 6: Review and practice
- Take Google’s official sample questions, linked from the exam page, and review every miss.
- Re-read the exam guide line by line and explain each bullet out loud in one sentence.
- Book the exam when no bullet makes you hesitate.
For hands-on practice, Google’s exam page points to the Google Cloud Free Tier, which gives free usage of select products up to monthly limits. Set a budget alert before you start any training jobs.
Turn old study notes into current ones
If you’ve got notes or flashcards from a Vertex AI-era course, don’t throw them away. Paste them into a chatbot with this prompt and the name table above:
You are helping me update my study notes for the Google Cloud Professional Machine Learning Engineer exam. The exam was updated for Gemini Enterprise Agent Platform (formerly Vertex AI), and the current exam guide is dated June 1, 2026.
Here is the official old-name to new-name mapping:
[PASTE THE NAME TABLE]
Here are my notes:
[PASTE NOTES OR FLASHCARDS]
Tasks:
1. Rewrite each note using the current product names. Keep my wording otherwise.
2. Flag any note about a topic that isn't in this list of current exam sections: [PASTE THE SIX SECTION HEADINGS AND BULLETS FROM THE EXAM GUIDE].
3. For each section, list any exam-guide bullet my notes don't cover at all.
Output a table with columns: Original note | Updated note | Still on the exam? (Yes / Not in guide / Unsure). If you're unsure whether a name changed, say so rather than guessing.Chatbots can mix up old and new names, so spot-check anything surprising against Google’s docs. Our guide to making AI more accurate and honest has a quick checking routine.
Common mistakes
- Studying an outdated blueprint. The older Vertex AI-era guide included TFX system design and hosting MLflow. The current outline drops them and adds Model Armor and Gemini tuning, so check any course against the June 2026 guide.
- Over-preparing code. Google says the exam doesn’t directly assess coding. Spend that time on service selection and trade-offs.
- Treating it like a data science exam. Serving, pipelines and monitoring add up to roughly half the weight. Production thinking matters as much as modeling.
- Ignoring the retake timeline. Under Google’s retake policy, you wait 14 days after a first failure, 60 days after a second and 365 days after a third, with a maximum of four attempts in two years. Each attempt is paid.
- A name mismatch on exam day. Your legal name in your certification account must match your government ID, or you can’t test and the fee is forfeited.
After you pass: renewal
Professional certifications last two years. For the Professional ML Engineer, renewal is by exam: the window opens 60 days before your certification’s inactive date, and passing adds two years. You can still renew up to 30 days after the inactive date. Google gives you a 50% off renewal discount code when you first certify, stored in your certification account profile. Our guide to which AI certifications expire compares this with other vendors.
Try it now
Download the current exam guide, highlight every bullet with a product you haven’t touched, and pick one to build this week. If you’re weighing this exam against AWS’s option, read Google ML Engineer vs AWS ML Engineer. Want an easier Google credential first? Start with our Google Generative AI Leader study guide, and see the full ranking of AI certifications for everything else.
🎁 Free download: The AI Cert Starter Kit
A one-page decision guide, a side-by-side comparison of the major AI certifications, and a 4-week study plan template, all in one free Notion doc. Free with your email address.
Related guides
- Free AI exam practice questions
- The Best AI Certifications in 2026, Ranked: Which Should You Get?
- Google Cloud Generative AI Leader Study Guide & Cheat Sheet
- AWS ML Engineer Associate (MLA-C01) Study Guide & Cheat Sheet
- Do AI Certifications Actually Increase Your Salary? An Honest Answer
HOW TO // AI is not affiliated with or endorsed by Google Cloud. Exam details come from Google’s official certification pages and exam guide as of October 2026 and can change. Check the official exam page before you book.
Get an email when the Google Cloud Certified Professional Machine Learning Engineer exam changes
We check the official exam pages every day. If the price, version, format or retirement date changes, you’ll get a short email. At most one a week, plus the HOW TO // AI newsletter on Thursdays.
Unsubscribe anytime. Privacy policy.
Frequently asked questions
Is the Google Professional Machine Learning Engineer exam hard?
It's a professional-level exam, and Google recommends 3+ years of industry experience including at least a year on Google Cloud. Google says it doesn't directly assess coding, so expect to be tested on choosing the right product and approach. Hands-on experience with training, serving and monitoring models matters more than memorizing facts.
Can I still study from Vertex AI materials?
Yes, as long as you translate the names. Google renamed Vertex AI products under Gemini Enterprise Agent Platform, and the current exam guide uses the new names. Check older courses against the current guide, because some topics were added and others removed.
What score do I need to pass?
Google doesn't publish a passing score. Its exams report only pass or fail, and if you fail you get a report showing how you did by section.
How long is the certification valid?
Two years. You can renew by taking the exam again in the window that opens 60 days before your certification's inactive date, and Google provides a 50% off renewal discount code when you first certify.
