GCP PMLE study guide · 2026

The 6 GCP PMLE domains, explained

6 domains and 14 official objectives. Scaling prototypes into ML models carries the most weight at ~21%. Each guide explains what the domain covers, how its questions are framed and where candidates lose marks.

Start with Domain 3 See all domains
Questions
50–60 (multiple choice)
Time
2 hours
Passing score
Not published by Google
Outline
Professional Machine Learning Engineer exam guide as of June 1, 2026 (current, Agent Platform product names)

Exam blueprint

Bar widths follow the official weights. Weight ranges are drawn at their midpoint.

  1. 1 13%
  2. 2 16%
  3. 3 21%
  4. 4 20%
  5. 5 18%
  6. 6 13%
  1. 1 Architecting low-code AI solutions
  2. 2 Collaborating within and across teams to manage data and models
  3. 3 Scaling prototypes into ML models
  4. 4 Serving and scaling models
  5. 5 Automating and orchestrating ML pipelines
  6. 6 Monitoring AI solutions

All 6 domains

14 official objectives 64–92 hrs suggested study

  1. Domain 01 ~13%

    Architecting low-code AI solutions

    Learn when a problem can be solved without custom training: BigQuery ML and AutoML for predictive models, prebuilt Google Cloud AI APIs for common tasks, and Gemini, Imagen, Veo and Model Garden for generative use cases.

    Inside this domain

    • Developing ML models using BigQuery ML or AutoML on Gemini Enterprise Agent Platform
    • Building AI solutions using Google Cloud AI APIs or foundational models

    Watch for Export to train

    Objectives
    2
    Questions est.
    ~7–8
    Study time est.
    8–12 hrs
  2. Domain 02 ~16%

    Collaborating within and across teams to manage data and models

    This section covers the shared work behind every model: exploring and preparing data with the right tool for its size, managing features and sensitive data, prototyping in secure notebooks, and tracking experiments, evaluations and lineage so teams can compare and reproduce results.

    Inside this domain

    • Exploring and preprocessing data for ML
    • Model prototyping using notebooks (e.g., Gemini Enterprise Agent Platform Workbench and Colab Enterprise)
    • Tracking and running ML experiments

    Watch for Pandas at scale

    Objectives
    3
    Questions est.
    ~8–10
    Study time est.
    10–14 hrs
  3. Domain 03 ~21%

    Scaling prototypes into ML models

    The largest section of the exam: choose a model type and product under cost and latency limits, train with custom jobs, AutoML or Kubeflow, tune hyperparameters, fine-tune foundation models when warranted, and pick CPUs, GPUs or TPUs for distributed training.

    Inside this domain

    • Building models given the task considering cost, complexity, latency, and scalability
    • Training models
    • Choosing appropriate hardware for training

    Watch for More GPUs for a slow job

    Objectives
    3
    Questions est.
    ~11–13
    Study time est.
    14–20 hrs
  4. Domain 04 ~20%

    Serving and scaling models

    Put trained models into production: choose batch or online inference, package models in prebuilt or custom containers, version them in Model Registry, roll out safely with canary and A/B tests, and scale endpoints for latency and throughput.

    Inside this domain

    • Serving models
    • Scaling online model serving

    Watch for Endpoint for nightly scoring

    Objectives
    2
    Questions est.
    ~10–12
    Study time est.
    12–18 hrs
  5. Domain 05 ~18%

    Automating and orchestrating ML pipelines

    Make ML repeatable: build end-to-end pipelines that validate data and models, choose an orchestrator such as Agent Platform Pipelines, Managed Service for Apache Airflow or Ray, keep preprocessing consistent, and automate retraining through CI/CD/CT.

    Inside this domain

    • Developing end-to-end ML pipelines
    • Automating model retraining

    Watch for Self-managed Kubeflow

    Objectives
    2
    Questions est.
    ~9–11
    Study time est.
    12–16 hrs
  6. Domain 06 ~13%

    Monitoring AI solutions

    Keep AI solutions safe and accurate after launch: block data leaks and malicious prompts with Model Armor and safety filters, apply responsible AI and explainability, and monitor production models for skew, drift and gen AI quality.

    Inside this domain

    • Identifying risks to AI solutions
    • Monitoring, testing, and troubleshooting AI solutions

    Watch for Retrain to fix skew

    Objectives
    2
    Questions est.
    ~7–8
    Study time est.
    8–12 hrs

Study the heaviest domains first

Marks follow weight. This order puts the most marks behind you first, across about 64–92 hrs of study.

  1. 3 Scaling prototypes into ML models ~21% · 14–20 hrs
  2. 4 Serving and scaling models ~20% · 12–18 hrs
  3. 5 Automating and orchestrating ML pipelines ~18% · 12–16 hrs
  4. 2 Collaborating within and across teams to manage data and models ~16% · 10–14 hrs
  5. 1 Architecting low-code AI solutions ~13% · 8–12 hrs
  6. 6 Monitoring AI solutions ~13% · 8–12 hrs

Try one

Practice questionDomain 3: Scaling prototypes into ML models

A support team wants a Gemini assistant to answer in a fixed JSON format with the company's tone. Prompt design with examples still gives inconsistent formatting across thousands of test cases. The team has 3,000 high-quality example pairs. What should the ML engineer do next?

One trap from every domain

The answer that looks right, and the clue that gives it away.

  • Domain 1

    Export to train

    Clue Data is already in BigQuery and the team writes SQL.

  • Domain 2

    Pandas at scale

    Clue Data size is in terabytes or growing daily.

  • Domain 3

    More GPUs for a slow job

    Clue Low accelerator utilization or a busy CPU.

  • Domain 4

    Endpoint for nightly scoring

    Clue A schedule and a large table to score.

  • Domain 5

    Self-managed Kubeflow

    Clue The team wants minimal infrastructure management.

  • Domain 6

    Retrain to fix skew

    Clue Serving inputs differ from training right after launch.

Other Artificial Intelligence & Machine Learning certifications

Study guides for exams that overlap with GCP PMLE.

Practise what you read

GCP PMLE domains: common questions

How many domains are in the GCP PMLE exam?

GCP PMLE has 6 domains with 14 official objectives between them: Architecting low-code AI solutions, Collaborating within and across teams to manage data and models, Scaling prototypes into ML models, Serving and scaling models, Automating and orchestrating ML pipelines and Monitoring AI solutions.

Which GCP PMLE domain carries the most weight?

Scaling prototypes into ML models is the heaviest at ~21%, followed by Serving and scaling models (~20%). The lightest is Monitoring AI solutions (~13%). Weights come from Google Cloud's official outline.

What is the GCP PMLE exam format?

Questions: 50–60 (multiple choice). Time allowed: 2 hours. Passing score: Not published by Google.

How long should I study for GCP PMLE?

Our planning estimate is 64–92 hrs across all domains, split by weight. It is an editorial estimate, not an official figure, so adjust it to your experience and to how you score on practice questions.

Where should I start?

Start with Scaling prototypes into ML models, because it carries the most marks, then work down by weight. Use the checklist on each domain page to confirm you can do what the domain asks before moving on.

Which outline is this guide based on?

This guide follows Professional Machine Learning Engineer exam guide as of June 1, 2026 (current, Agent Platform product names), published by Google Cloud. It was last reviewed in October 2026.

Domain names, weights and objectives come from Google Cloud's official outline (Professional Machine Learning Engineer exam guide as of June 1, 2026 (current, Agent Platform product names)). Question counts and study hours are editorial planning estimates, not official figures. Last reviewed October 2026.