Updated for MLA-C02. MLA-C01 retired in English on September 28, 2026.

See the full timeline
AssociateMLA-C02

AWS Machine Learning Engineer Associate (MLA-C02) Practice Exam

Validate your ability to build, operationalize, deploy, and maintain ML solutions and pipelines on AWS. Covers data preparation, model development, deployment orchestration, and ML solution monitoring and security.

65 questions
2h 10m
72% to pass

No sign-up — instant feedback after every question

⚡

Short on time?

10 questions, about 10 minutes, instant feedback. No timer, no signup.

Start a 10-minute drill

Practice Mode

Get instant feedback after each question. See explanations and track your progress as you learn.

Exam Mode

Timed session with no feedback until the end. Experience real exam conditions and test your knowledge.

Strike out answers you have ruled out, like on the real exam.

Free cheat sheet13 pages PDF

AWS Certified Machine Learning Engineer – Associate Cheat Sheet (MLA-C02)

Every in-scope service and decision the MLA-C02 exam tests — classic SageMaker AI engineering plus Bedrock, RAG, agents and generative AI evaluation, organised by the four exam domains with closed services tagged

About the AWS Machine Learning Engineer Associate (MLA-C02) Certification

The AWS Certified Machine Learning Engineer Associate (MLA-C02) is AWS's certification for practitioners who build, deploy, and maintain ML and AI solutions. MLA-C02 is in beta in English (MLA-C01 retired in English on September 28, 2026, and stays available in Japanese, Korean and Simplified Chinese until MLA-C02's general availability, which AWS's September 2026 announcement set for January 14, 2027; the AWS exam page now lists the GA date as TBD). It keeps the practical engineering focus (data pipelines, model training, deployment, monitoring and MLOps on SageMaker AI) while adding foundation models, Amazon Bedrock, RAG and agentic workflows. This practice exam is written for MLA-C02 and covers all four domains, including Amazon Bedrock, RAG, agents and generative AI evaluation.

This exam targets ML engineers, data engineers, and software engineers who build and deploy ML models on AWS. Unlike the AI Practitioner (which is conceptual), this certification requires hands-on experience with SageMaker AI, Amazon Bedrock, data pipelines, and model deployment. It's a solid credential for anyone transitioning into ML engineering from a software or data background.

Read the full AWS Machine Learning Engineer Associate certification guide

Exam Format and Details

The MLA-C02 beta exam has 85 questions in multiple-choice and multiple-response format, and you have 170 minutes to complete it. The exam guide describes the standard exam as 65 questions (50 scored, 15 unscored). Results are reported on a scaled score of 100–1,000, and the passing score is 720. The beta costs $75 USD and is English only. MLA-C01 (65 questions in 130 minutes) retired in English on September 28, 2026.

Key Topics Covered

  • Data Preparation for ML and AI (28% of exam)
  • ML Model and Foundation Model (FM) Development (24% of exam)
  • Deployment and Orchestration of ML and AI Workflows (24% of exam)
  • Operating, Monitoring, and Securing ML and AI Solutions (24% of exam)

Recommended experience: At least one year of experience using Amazon SageMaker AI, Amazon Bedrock and other AWS services for ML engineering, plus a year in a related role such as backend developer, DevOps engineer, data engineer or data scientist, with experience of both traditional ML and generative AI.

Try 3 sample MLA-C02 questions

Written in the style of the real exam. Answer each one, then open the answer to check.

  1. Question 1 · Data Preparation for ML and AI

    A company is building a retrieval augmented generation (RAG) assistant with Amazon Bedrock Knowledge Bases over a library of long technical manuals. Testing shows that searches find the right passage, but the retrieved chunks are so small that the model's answers miss details from the rest of the same section. The company wants to keep precise matching on small passages while giving the model the broader context around each match. Which chunking strategy should the ML engineer configure for the data source?

    • A.Fixed-size chunking with a smaller maximum number of tokens per chunk
    • B.No chunking, so that each manual is stored as a single chunk
    • C.Fixed-size chunking with a larger overlap percentage between chunks
    • D.Hierarchical chunking with small child chunks and large parent chunks
    Show answer

    Answer: D

    Option D is correct: hierarchical chunking searches against small child chunks for precise matches and then replaces them with their larger parent chunks in the results, so the model receives the surrounding context. Option A makes each chunk even smaller, which removes more context. Option B gives the model whole manuals with imprecise matching, because one embedding has to represent an entire document. Option C repeats a few tokens between neighboring chunks but does not add the section-level context the answers are missing.

    Sources: AWS documentation

  2. Question 2 · ML Model and Foundation Model (FM) Development

    An ML engineer trains a computer vision model with an Amazon SageMaker AI training job on a single GPU instance. The training data is 4 TB of image files in Amazon S3. The job uses File input mode, and each run spends a long time copying data to the instance before the first epoch starts. The engineer wants to reduce this startup time without changing how the training code reads files. Which actions will meet these requirements? (Select TWO.)

    • A.Increase the size of the ML storage volume that is attached to the training instance.
    • B.Change the input mode of the training channel to FastFile mode.
    • C.Turn on network isolation for the training job so that data transfer stays inside AWS.
    • D.Set the S3 data distribution type to ShardedByS3Key while keeping File input mode.
    • E.Use an Amazon FSx for Lustre file system linked to the S3 bucket as the data source.
    Show answer

    Answers: B, E

    Options B and E are correct: FastFile mode exposes the S3 objects as files that are streamed on demand, so training starts without waiting for a full copy, and FSx for Lustre linked to the bucket mounts the data as a high-throughput file system that the code reads with normal file paths. Option A makes room for the data but the full download still happens before training starts. Option C restricts network access for security and does not speed up data loading. Option D only splits the data across multiple instances, so with a single instance the whole dataset is still downloaded.

    Sources: AWS documentation

  3. Question 3 · Deployment and Orchestration of ML and AI Workflows

    A company has trained a small scikit-learn model that classifies support tickets. An internal application calls the model synchronously a few hundred times each day in short bursts, with idle periods of several hours between bursts. Users can accept a few extra seconds of latency on the first request after an idle period. Which deployment option will meet these requirements MOST cost-effectively?

    • A.Deploy the model to an Amazon SageMaker AI serverless inference endpoint.
    • B.Deploy the model to a real-time endpoint that runs on one ml.m5.large instance.
    • C.Start a SageMaker AI batch transform job each time the application sends a request.
    • D.Deploy the model to an asynchronous inference endpoint that scales down to zero.
    Show answer

    Answer: A

    Option A is correct: serverless inference scales to zero when there is no traffic and bills only for the compute used to process requests, and it is designed for intermittent traffic that can tolerate cold starts. Option B bills for the instance around the clock, including the long idle periods. Option C has startup times measured in minutes and is meant for offline datasets, not synchronous requests. Option D can scale to zero, but it queues requests and writes results to Amazon S3, so the application would not get a synchronous response.

    Sources: AWS documentation 1, AWS documentation 2

These sample questions are not part of the MLA-C02 practice exam, so you won't meet them again when you take it.

Watch: AWS Machine Learning Engineer Associate Exam Preparation

Frequently Asked Questions

How many questions are on this practice exam, and how long do I get?

This practice exam has 65 questions, and exam mode gives you 130 minutes to answer them. That is the number of questions the AWS exam guide gives for the standard MLA-C02 exam (50 scored and 15 unscored). The MLA-C02 beta you can book now is longer: 85 questions in 170 minutes. AWS has not yet published the time limit for the standard exam. Practice mode has no timer.

Is this practice exam like the real MLA-C02 exam?

In format, yes: questions are multiple choice (one correct answer) or multiple response (two or more), spread across the 4 exam domains in close to the weights the official exam guide gives them. Scoring differs: AWS reports a scaled score from 100 to 1,000 and you pass at 720, while this practice exam shows the raw percentage you got right. For how hard the exam is, whether it is worth it and how long to study, read our AWS Machine Learning Engineer Associate certification guide.

What score should I aim for before booking the real exam?

Aim for 80% or more, consistently, on full-length attempts in exam mode, not just once. That leaves a margin over the 720 out of 1,000 you need on the real MLA-C02 exam, whose questions will all be new to you. Repeat attempts here show the same questions, so judge yourself by first attempts and by whether you can explain each answer. No practice score guarantees a pass.

Do I need to sign up or pay?

No. The MLA-C02 practice exam is free and needs no account. Your progress and score history are saved in this browser, so they stay on this device and are lost if you clear your browser data.

How should I use this practice exam?

Start in practice mode, which tells you after each question whether you were right and explains the answer. When you score well there, switch to exam mode: 130 minutes and no feedback until the end, like test day. Your results break the score down by domain, so you can see which domains to study before your next attempt.

Is this practice exam for MLA-C01 or MLA-C02?

MLA-C02. The last day to take MLA-C01 in English was September 28, 2026, and the MLA-C02 beta started on September 29, 2026. MLA-C01 stays available in Japanese, Korean and Simplified Chinese until MLA-C02 becomes generally available. This practice exam follows the MLA-C02 exam guide, which adds foundation models, RAG and agentic workflows to the data preparation, training, deployment and MLOps topics.

Affiliate links — if you enrol through them, CloudNinja may earn a commission at no extra cost to you.

Studying for more than one certification? All our practice tests are free. For a daily warm-up, try the AWS Question of the Day.