AWS Certified Machine Learning Engineer - Associate (MLA-C01) is the Associate-level certification that goes beyond training ML models with SageMaker AI to cover the whole job of deploying, automating, monitoring, and securing them. Since September 29, 2026 the English exam has been the revised MLA-C02 (beta), while MLA-C01 in Korean, Japanese, and Simplified Chinese remains available until MLA-C02 reaches general availability, so this guide is organized around MLA-C01 and points out what has changed.
This post is the guide to our 13-part MLA-C01 domain deep-dive series. Before getting into individual topics, it maps out the exam structure, who should take it, and where this certification sits on the AWS career path.
---
Who Is MLA-C01 For?
MLA-C01 resembles building and running a factory. Even with a good blueprint (the model), the factory only runs if you bring in raw materials (data), automate the production line (pipelines), watch the defect rate (drift), and control who gets in (security). The exam likewise asks less about model accuracy and more about how you assemble that whole operation from AWS services.
This certification is a particularly good fit for the following people.
| Target Audience | Why It Fits | |------|------| | ML engineers and MLOps engineers | Formally validates the hands-on skills of training, deploying, and automating with SageMaker AI | | Data scientists | Adds the deployment, monitoring, and cost sense needed to take models out of the notebook into production | | Backend and DevOps engineers | Extends familiar CI/CD, IaC, and security knowledge to ML workloads | | Data engineers | Extends data pipelines all the way to model training and inference | | AIF-C01 holders | The next step up from understanding concepts to actually building and operating |
There are no prerequisite certifications. AWS recommends at least one year of ML engineering experience with SageMaker AI and related services, plus at least one year in a related role such as backend development, DevOps, data engineering, or data science. What matters most is not ML theory but judgment in choosing SageMaker AI features and combinations of AWS services.
---
Exam at a Glance
| Item | Detail | |------|------| | Exam Code | MLA-C01 | | Full Name | AWS Certified Machine Learning Engineer - Associate | | Level | Associate | | Passing Score | 720 (scale 100–1,000) | | Questions | 65 (50 scored + 15 unscored) | | Duration | 130 minutes | | Exam Fee | $150 USD | | Validity | 3 years | | Question Types | Multiple choice, multiple response, ordering, matching, case study | | Languages | Korean, Japanese, Simplified Chinese (MLA-C01 in English ended on September 28, 2026) | | Status | Being revised: English is now the MLA-C02 beta (170 minutes · 85 questions · $75 USD) | | Delivery | Pearson VUE testing center or online proctored exam | | Recommended Experience | 1 year of ML engineering with SageMaker AI and more + 1 year in a related role (not required) |
The 15 unscored questions are there so AWS can evaluate them for future exams, and they are not identified. Unanswered questions are scored as incorrect and there is no penalty for guessing. Scoring is compensatory, so there is no per-domain cutoff; you only need 720 overall. Besides multiple choice and multiple response, there are ordering, matching, and case study questions; multiple response, ordering, and matching give credit only when everything is correct, while each case study question is scored separately.
---
Domain Breakdown
MLA-C01 is made up of four domains. The weights below are the official figures for scored questions.
| Domain | Topic | Weight | Deep dives | |------|------|------|------| | Domain 1 | Data Preparation for ML | 28% | 3 posts | | Domain 2 | ML Model Development | 26% | 5 posts | | Domain 3 | Deployment and Orchestration of ML Workflows | 22% | 3 posts | | Domain 4 | ML Solution Monitoring, Maintenance, and Security | 24% | 2 posts |
Domain 1 (data preparation) is the largest at 28%, but the other three are evenly spread between 22% and 26%. Going deep in only one domain will not get you past the passing score; you need to study the whole flow evenly, from data preparation to model development, deployment, and operations. In our question bank, endpoint scaling, security, managed AI services, and streaming and event automation come up especially often.
!MLA-C01 exam domain weight breakdown across 4 domains
Domain 1: Data Preparation for ML (28%)
This domain is about collecting data, turning it into a form a model can learn from, and checking its quality and bias.
Key topics: Choosing storage such as S3, EFS, and FSx and formats such as Parquet, ORC, and RecordIO; streaming ingestion with Kinesis, Firehose, Managed Service for Apache Flink, and MSK; transformation with Glue, DataBrew, Data Wrangler, and EMR; feature engineering such as scaling, encoding, and handling missing values; Feature Store; Ground Truth labeling; Glue Data Quality and Clarify pre-training bias metrics; encryption, masking, and PII and PHI compliance.
Exam tip: "Training and inference must use the same feature values" → Feature Store. "Clean data quality without code" → Glue DataBrew. "Large distributed training needs a fast shared file system" → FSx for Lustre.
Related posts: Data Ingestion and Storage, Streaming and Event-Driven ML Pipelines, Feature Engineering and Data Quality
Domain 2: ML Model Development (26%)
This domain is about choosing the right modeling approach for the problem, training and tuning, and evaluating performance.
Key topics: Choosing among managed AI services, SageMaker AI built-in algorithms, JumpStart, and Bedrock; Script Mode, distributed training, and Spot training; automatic model tuning (Bayesian, random, Hyperband); regularization, early stopping, and ensembles; catastrophic forgetting; reducing model size; the model registry; evaluation metrics such as the confusion matrix, F1, AUC, and RMSE; Clarify and Debugger; comparing shadow variants.
Exam tip: "Fix product names that keep being transcribed wrong, without ML specialists" → Transcribe custom vocabulary. "High training accuracy, low validation accuracy" → overfitting, so regularization and early stopping. "99% accuracy on data where 0.5% is fraud" → evaluate with recall and F1.
Related posts: Choosing Modeling Approaches, Choosing AWS AI Services, ML Fundamentals for Cloud Engineers, Model Training, Tuning, and Evaluation, Evaluation Metrics and Class Imbalance
Domain 3: Deployment and Orchestration of ML Workflows (22%)
This domain is about putting trained models into service, scaling them, and automating the whole process.
Key topics: Choosing real-time, serverless, asynchronous, or batch inference; multi-model and multi-container endpoints; choosing the deployment target among SageMaker endpoints, EKS, ECS, and Lambda; auto scaling policies and metrics; CloudFormation and the CDK; containers (ECR, BYOC); endpoints inside a VPC; SageMaker Pipelines, Step Functions, MWAA, and EventBridge; CodePipeline, CodeBuild, and CodeDeploy with blue/green, canary, and linear deployments.
Exam tip: "Prepare for the 9 a.m. peak in advance and handle other swings too" → scheduled scaling + target tracking. "Process large payloads for minutes and scale to zero when idle" → asynchronous inference. "Deploy automatically once a model is approved" → an EventBridge rule on the model registry approval event.
Related posts: Deployment Infrastructure and Scaling Strategies, Endpoint Auto Scaling and Deployment Targets, Building MLOps and CI/CD Pipelines
Domain 4: ML Solution Monitoring, Maintenance, and Security (24%)
This domain is about watching models and infrastructure in production, cutting costs, and protecting the system.
Key topics: Watching data and model quality drift with Model Monitor; A/B testing; observing and auditing with CloudWatch, X-Ray, and CloudTrail; rightsizing with Inference Recommender and Compute Optimizer; managing costs with Spot, Savings Plans, tags, and Budgets; IAM least privilege; VPC and network isolation; KMS encryption; CI/CD security.
Exam tip: "The input data distribution in production differs from training" → Model Monitor data quality. "Training data must not traverse the internet" → private subnets + an S3 gateway endpoint. "Who changed the endpoint configuration?" → CloudTrail.
Related posts: Model Monitoring, Cost Optimization, and Security, ML Security Deep Dive
---
MLA-C01 vs. MLA-C02, and the AWS AI/ML Certification Path
AWS AI/ML certifications run from Foundational (AIF-C01) to Associate (MLA) to Professional (AIP-C01), and MLA is moving to a revised version in fall 2026. Start with the differences between the two versions.
| Category | MLA-C01 (current) | MLA-C02 (revised beta) | |------|------|------| | Languages | Korean, Japanese, Simplified Chinese (English ended on September 28, 2026) | English only (Korean, Japanese, and Simplified Chinese planned at general availability) | | Exam Format | 130 minutes · 65 questions · $150 USD | 170 minutes · 85 questions · $75 USD (beta pricing) | | Question Types | Multiple choice, multiple response, ordering, matching, case study | Multiple choice, multiple response | | Domain Weights | 28% · 26% · 22% · 24% | 28% · 24% · 24% · 24% | | Scope Changes | The baseline scope of this series | Adds Bedrock knowledge bases, RAG, embeddings, agents, and generative AI evaluation; drops Neo, BYOC, and model size reduction |
Whichever exam you pass, you earn the same AWS Certified Machine Learning Engineer - Associate certification, valid for three years. If you test in Korean, prepare with this series as it is; if you test in English, add the generative AI scope with the AIP-C01 guide. The former higher-level certification, Machine Learning - Specialty (MLS-C01), was retired on March 31, 2026, and the next AI/ML step today is the Professional-level AIP-C01 (180