Google Cloud PMLE 2026 考試指南：形式、六個範疇與備試計劃

Google Cloud has updated its Professional Machine Learning Engineer (PMLE) exam, and the current exam guide is dated 1 June 2026. Here is the format, the six exam guide sections and their weights, the recommended experience, how renewal works, and a study plan.

What changed

Google’s certification 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 that it now prioritises Google Cloud native solutions. The June 2026 guide uses the new product names throughout, such as Agent Platform AutoML, Agent Platform Pipelines and Agent Platform Workbench. If your study material still says Vertex AI, check it against the new guide.

The exam at a glance

Length: two hours
Format: 50–60 multiple choice and multiple select questions
Fee: USD 200, plus tax where applicable
Languages: English and Japanese
Delivery: online-proctored from a remote location, or onsite-proctored at a testing centre
Results: pass or fail only, with no numerical score. You see a provisional result on screen as soon as you submit
Coding: the exam does not directly assess coding skills, but you need a minimum proficiency in Python and SQL to interpret questions with code snippets

The six exam guide sections

Google publishes an approximate weight for each section:

Section	Weight
Architecting low-code AI solutions	~13%
Collaborating within and across teams to manage data and models	~16%
Scaling prototypes into ML models	~21%
Serving and scaling models	~20%
Automating and orchestrating ML pipelines	~18%
Monitoring AI solutions	~13%

Prerequisites and recommended experience

There are no prerequisites. Google recommends 3 or more years of industry experience, including at least 1 year designing and managing solutions using Google Cloud.

Renewal and retakes

Validity: Professional certifications are valid for two years from the date you certify
Renewal: you can recertify from 60 days before your certification expires. For the PMLE, renewal is by taking the exam again, and a pass adds two years
Retakes: up to four attempts in two years. After a first fail you wait 14 days, after a second 60 days, and after a third 365 days

A study plan by weight

The two middle sections, building models and serving them, make up about 41% of the exam. If you have about 40 hours, a split that follows the weights is roughly:

9 hours: Scaling prototypes into ML models (choosing model types and products, custom training, hyperparameter tuning, fine-tuning foundation models, CPU, GPU and TPU choices)
8 hours: Serving and scaling models (batch and online inference, Model Registry, A/B and canary rollouts, public and private endpoints)
7 hours: Automating and orchestrating ML pipelines (Agent Platform Pipelines, consistent preprocessing, retraining policies, CI/CD/CT with Cloud Build)
6 hours: Collaborating to manage data and models (preprocessing with BigQuery, Dataflow or Spark, Feature Store, handling PII, notebooks, experiment tracking)
5 hours: Architecting low-code AI solutions (BigQuery ML, Agent Platform AutoML, Model Garden, Document AI, Vision and Translate APIs)
5 hours: Monitoring AI solutions (Model Armor and safety filters, bias, explainability, training-serving skew and drift)

Google’s page suggests reviewing its sample questions and following the Machine Learning Engineer learning path. The Google Cloud Free Tier gives free usage, up to monthly limits, of selected products for hands-on practice.

Confirm the latest details on Google Cloud’s Professional Machine Learning Engineer page before you book.

I run ExamCrest, where this post first appeared. If you're preparing for PMLE, there are 10 free PMLE practice questions, each with the answer and a full explanation.