June 2026 PMLE Exam Update | WiseOwlLearns
The June 2026 PMLE replaces Vertex AI with Gemini Enterprise Agent Platform and shifts from recall to judgment. Learn what changed and how to prepare.
TL;DR: The June 2026 PMLE exam update completely replaces “Vertex AI” with “Gemini Enterprise Agent Platform”. The exam no longer tests rote memorization of algorithms; it now tests architectural judgment under constraint (cost, latency, scale). Old practice tests are now obsolete.
The shift to the Gemini Enterprise Agent Platform marks the most significant operational and pedagogical change in the history of the Google Cloud Professional Machine Learning Engineer (PMLE) certification.
If you are preparing for the exam after June 1, 2026, using outdated practice materials is a liability.
1. The Shifting Weights: Less Pipelines, More Judgment
Google has fundamentally rebalanced the exam to reflect modern AI workloads. The days of spending 25% of your study time on ML pipeline orchestration are over. The new exam heavily favors scaling prototypes and foundation models.
Here is the complete breakdown of how the exam weights have shifted across all six objectives:
| Objective | Pre-2026 | June 2026 | Shift |
|---|---|---|---|
| 1. Architecting low-code AI solutions | ~13% | ~13% | — |
| 2. Collaborating to manage data and models | ~16% | ~16% | — |
| 3. Scaling prototypes into ML models | ~18% | ~21% | 📈 +3% |
| 4. Serving and scaling models | ~20% | ~20% | — |
| 5. Automating and orchestrating ML pipelines | ~22% | ~18% | 📉 -4% |
| 6. Monitoring AI solutions | ~11% | ~13% | 📈 +2% |
The two most dramatic changes reveal Google’s updated priorities:
Major Shifts: Pre-2026 vs June 2026
Objective 3: Scaling Prototypes into ML Models (18% ➔ 21%)
Why the increase? The new exam emphasizes deploying foundation models, choosing between Cloud Run and Agent Platform Inference under latency constraints, and tuning LLMs.
Objective 5: Automating & Orchestrating ML Pipelines (22% ➔ 18%)
Why the decrease? Google has deprecated TFX and shifted focus away from low-level MLOps orchestration in favor of managed solutions like Managed Service for Apache Airflow.
Insight: The 21% weight on scaling prototypes means you must understand how to evaluate foundation models, when to use parameter-efficient fine-tuning (PEFT), and how to deploy to Agent Platform Inference under latency constraints.
2. The Rename: Vertex AI name is deprecated.
You cannot pass this exam if you don’t know the current nomenclature. As of June 1, 2026, the term “Vertex AI” is entirely deprecated.
| Old Term (Pre-June 2026) | New Canonical Term (June 2026+) |
|---|---|
| Vertex AI | Gemini Enterprise Agent Platform (GEAP) |
| Vertex AI Prediction | Agent Platform Inference |
| Vertex AI Workbench | Agent Platform Workbench |
| Vertex AI Pipelines | Agent Platform Pipelines |
| Cloud Composer | Managed Service for Apache Airflow |
🚨 The Vertex AI Distractor: If you see “Vertex AI Prediction” as an option on the actual exam, you can safely assume it is a distractor. The canonical service is now Agent Platform Inference.
3. Shift to Judgment Under Constraint
The previous iteration heavily tested your ability to recall which algorithm fit which data type. The new update explicitly frames Objective 3.1 as building models “considering cost, complexity, latency, and scalability.”
| Paradigm | Exam Focus |
|---|---|
| Old Exam (Pre-2026) | Memorizing which algorithm to use for structured data. |
| New Exam (2026+) | Choosing between Cloud Run and Agent Platform Inference under strict latency constraints. |
How to Decide on Deployment (Exam Logic)
- Needs Sub-millisecond Latency? → Custom Container on GKE / Cloud Run
- No strict latency? → Agent Platform Inference
- Is it a Foundation Model? → Model Garden API
- Is it a custom model? → Standard Endpoint
4. Generative AI & Security are Everywhere
Gen AI is no longer a separate, bolt-on topic; it is integrated throughout the ML lifecycle.
- Fine-Tuning: The exam tests your ability to fine-tune Gemini models directly using BigQuery.
- Evaluation: You must know how to evaluate Gen AI solutions using LLM-as-a-judge.
- Security: Objective 6.1 now lists specific Gen AI threats like data exfiltration and malicious prompting.
🛡️ Security Focus: You must understand how to implement Model Armor (a newly named product) to defend against prompt injection and enforce safety filters.
5. Why We Rebuilt WiseOwlLearns for the 2026 Exam
When Google announced these changes, we knew that simply finding and replacing “Vertex AI” with “Agent Platform” in existing question banks wouldn’t cut it. The shift required a fundamentally new approach.
Our platform includes the Option Analyzer™, designed to train your judgment under constraint. It walks you through expert elimination reasoning, evaluating each choice against data volume, latency, cost, and vendor lock-in.
Additionally, our WiseOwl Tutor™ acts as your interactive guide, trained on the most current 2026 GCP documentation.