Answer in brief
AWS Certified Machine Learning Engineer – Associate (MLA-C02) is an associate-level exam on building, deploying, operating and securing ML and AI solutions on AWS. It covers traditional ML models and foundation models (FMs). It is in beta: 85 questions in 170 minutes for $75, in English only. The pass mark is a scaled 720 out of 1,000.
- What it replaced: MLA-C01. The last day for MLA-C01 in English was 28 September 2026. MLA-C02 beta delivery began on 29 September 2026.
- What it tests: the full lifecycle. Prepare data, build and customise models, deploy and orchestrate them, then monitor, cost-manage and secure them.
- Domains: four. Data Preparation for ML and AI is the largest at 28%. The other three are 24% each.
- What is new: generative AI in every domain. That means Amazon Bedrock, RAG, vector databases, knowledge bases, agents, prompt management and Bedrock Guardrails.
- How to prepare: read the official exam guide, take a diagnostic by domain, fix the heaviest weak domains first, build one small project that covers both SageMaker AI and Bedrock, and book when practice results are strong in every domain.
Everything in this guide comes from the official AWS exam guide, the AWS certification page and the AWS beta exam policy unless I say otherwise.
See the study system in action
How StudyTech AI works
Watch how your answers become a focused daily plan, targeted practice and flashcards. See how StudyTech AI tracks your progress before you decide when to book.
Ready to explore? Visit StudyTech AI to see supported certifications and choose your next step.
Explore StudyTech AIThe exam at a glance
Beta: 85 questions, 170 minutes, $75, 720 to pass, no prerequisites.
| Detail | AWS Certified Machine Learning Engineer – Associate |
|---|---|
| Exam code | MLA-C02 (replaces MLA-C01) |
| Level | Associate |
| Status | Beta. Delivery began 29 September 2026. General availability date not yet announced |
| Questions | 85 (beta). The exam guide lists 50 scored and 15 unscored questions |
| Question types | Multiple choice (one correct of four) and multiple response (two or more correct of five or more) |
| Time | 170 minutes (beta) |
| Price | $75 USD (beta pricing) |
| Passing score | 720 on a scaled range of 100–1,000 |
| Scoring model | Compensatory: you pass the exam overall, not each domain |
| Delivery | Pearson VUE test centre or online proctored |
| Languages | English only during beta |
| Prerequisites | None required |
| Validity | 3 years |
Values in bold are beta values. AWS can change them at general availability, so check the official certification page before you book.
Four details worth knowing before you book:
- A beta pass is a full pass. AWS says candidates who pass a beta exam earn the certification, active for three years.
- Results take up to 5 business days. AWS says all exam results, including beta results, are available within 5 business days.
- Unscored questions are hidden. You cannot tell which questions count. Treat every question as scored.
- Blank answers are wrong. There is no penalty for guessing, so never leave a question empty.
MLA-C01 vs MLA-C02: what changed
Same four domains and task statements, new generative AI skills in every domain, a few old topics removed, and two weights moved.
AWS publishes an official comparison. Here is what it says.
| Domain | MLA-C01 | MLA-C02 |
|---|---|---|
| 1. Data Preparation | 28% (Data Preparation for Machine Learning (ML)) | 28% (Data Preparation for ML and AI) |
| 2. Model Development | 26% (ML Model Development) | 24% (ML Model and Foundation Model (FM) Development) |
| 3. Deployment and Orchestration | 22% (Deployment and Orchestration of ML Workflows) | 24% (Deployment and Orchestration of ML and AI Workflows) |
| 4. Monitoring and Security | 24% (ML Solution Monitoring, Maintenance, and Security) | 24% (Operating, Monitoring, and Securing ML and AI Solutions) |
What AWS added, in short:
- Domain 1: scalable vector databases (OpenSearch Service, Amazon RDS with pgvector, Amazon S3), multimodal data, embedding models, text pre-processing, RAG document preparation (chunking, metadata extraction), masking and redaction, data for FM fine-tuning and distillation, and training data integrity checks.
- Domain 2: choosing FMs on Amazon Bedrock, fine-tuning strategies, RAG architecture patterns, performance/latency/cost trade-offs, prompt engineering, retrieval and embedding optimisation, human evaluation, NLP metrics (BLEU, ROUGE, BERTScore), LLM-as-a-judge and RAG monitoring.
- Domain 3: FM deployment options, Amazon Bedrock Custom Model Import, agent deployment and communication protocols, reranking, Bedrock knowledge bases, agent state management, GPU scaling, Amazon Bedrock Prompt Management, agent version management, prompt testing and knowledge base refresh pipelines.
- Domain 4: agent monitoring (tool failures, truncated streaming), Amazon Bedrock evaluations, FM inference and token costs, CI/CD vulnerability scanning (Amazon CodeGuru, Amazon Inspector), Bedrock API keys versus IAM credentials, and Amazon Bedrock Guardrails.
What AWS removed:
- Configuring data to load into training resources (Amazon EFS, Amazon FSx)
- Using custom datasets to fine-tune pre-trained models, as written in MLA-C01 (fine-tuning now appears in new, broader skills)
- Reducing model size (pruning, compression and similar)
- Optimising models on edge devices (SageMaker Neo)
- Bring your own container (BYOC) with SageMaker
- Monitoring infrastructure with Amazon EventBridge events, and troubleshooting capacity concerns such as provisioned concurrency and service quotas
The format changed too. MLA-C01 is 65 questions in 130 minutes for $150. The MLA-C02 beta is 85 questions in 170 minutes for $75.
If your course or practice test was built for MLA-C01, it will miss the generative AI skills and spend time on removed topics.
The four domains and their weights
Domain 1 is the largest at 28%; the other three are equal at 24%.
| # | Domain | Weight | Rough scored questions (of 50) |
|---|---|---|---|
| 1 | Data Preparation for ML and AI | 28% | ~14 |
| 2 | ML Model and Foundation Model (FM) Development | 24% | ~12 |
| 3 | Deployment and Orchestration of ML and AI Workflows | 24% | ~12 |
| 4 | Operating, Monitoring, and Securing ML and AI Solutions | 24% | ~12 |
The weights are AWS's. The "rough scored questions" column is my own arithmetic (weight × 50 scored questions from the exam guide), not an AWS figure. Use it to plan effort, not as a count.
The services named below are the examples AWS gives in each domain's skill list. AWS says the guide is not a complete list of exam content, so treat them as the core, not the edge.
Domain 1: Data Preparation for ML and AI (28%)
What it tests: getting data in, shaping it, and proving it is fit to use. Three tasks:
- Collect and store data
- Perform data transformation, feature engineering, and pre-processing
- Validate data quality and manage bias
Services that matter: Amazon S3, Amazon EBS, Amazon EFS, Amazon RDS (including pgvector), Amazon DynamoDB, Amazon OpenSearch Service, Amazon Kinesis, AWS Glue, AWS Glue DataBrew, AWS Glue Data Quality, Spark on Amazon EMR, AWS Lambda, SageMaker Data Wrangler and SageMaker Feature Store. Apache Flink, Apache Kafka and Apache Spark are named too.
Expect data formats (Parquet, JSON, CSV, ORC), streaming ingestion, feature engineering (scaling, binning, normalisation), class imbalance and bias. The new material is RAG preparation: chunking, embeddings, vector stores, masking sensitive data and preparing data for fine-tuning.
Domain 2: ML Model and Foundation Model (FM) Development (24%)
What it tests: choosing the right model approach, training or customising it, and evaluating it. Three tasks:
- Choose appropriate modeling approaches for ML and AI solutions
- Train, fine-tune, and customize models for ML and AI solutions
- Analyze and evaluate the performance of ML and AI systems
Services that matter: Amazon Bedrock (FM selection, Amazon Bedrock evaluations, Amazon Bedrock Prompt Management), Amazon SageMaker AI (built-in algorithms, script mode, automatic model tuning), MLflow on SageMaker AI, and the AWS AI services Amazon Textract, Amazon Rekognition, Amazon Comprehend and Amazon Transcribe.
Expect when to use a managed AI service versus a custom model, fine-tuning versus RAG, hyperparameters, overfitting and catastrophic forgetting, drift baselines, shadow variants, and GenAI evaluation (BLEU, ROUGE, BERTScore, human-in-the-loop, LLM-as-a-judge).
Domain 3: Deployment and Orchestration of ML and AI Workflows (24%)
What it tests: getting models and agents into production and automating the path there. Three tasks:
- Manage deployment infrastructure for ML and AI model types
- Provision and configure resources for ML and AI workloads based on existing architecture and requirements
- Implement automated orchestration and continuous integration and continuous delivery (CI/CD) pipelines for MLOps and AI workloads
Services that matter: SageMaker AI endpoints (including in a VPC), the SageMaker AI SDK for Python, AWS CLI and Boto3, Amazon Bedrock Custom Model Import, Amazon Bedrock knowledge bases, Amazon Bedrock Prompt Management, SageMaker Model Registry, MLflow on SageMaker AI, and AWS CodeBuild, AWS CodeCommit, AWS CodeDeploy, AWS CodePipeline and AWS CodeConnections.
Expect real-time versus batch inference, on-demand versus provisioned resources, auto scaling metrics, containers, rollback strategies and retraining. The new material is agents: deploying them, managing their state and versions, plus reranking, retrieval pipelines and knowledge base refresh cycles.
Domain 4: Operating, Monitoring, and Securing ML and AI Solutions (24%)
What it tests: keeping models and agents healthy, affordable and secure once they are live. Three tasks:
- Monitor ML and AI model inference and performance
- Optimize and manage ML and AI infrastructure costs and performance
- Secure ML and AI workloads and model endpoints
Services that matter: Amazon CloudWatch (including generative AI observability), Amazon Bedrock Model Evaluation, Amazon Bedrock AgentCore Observability, AWS X-Ray, Amazon CodeGuru, Amazon Inspector, IAM, AWS CloudTrail, AWS Config, Amazon Bedrock API keys and Amazon Bedrock Guardrails. VPCs, subnets and security groups appear too.
Expect drift and data distribution changes, A/B testing, instance families and purchasing options, cost quotas, and least privilege. The new material is agent monitoring (tool failures, coordination failures), token and vector storage costs, and Guardrails for sensitive data and responsible AI.
What is out of scope
AWS lists four job tasks the target candidate is not expected to do:
- Designing and architecting full end-to-end AI and ML solutions
- Setting up best practices and guiding ML strategies
- Handling integration with a wide array of services or new tools and technologies
- Working deeply in two or more ML domains
This is an engineer's exam, not an architect's. Questions ask you to pick and configure the right option inside a given setup.
Who should take it
People who already work with data, ML or backend systems on AWS and want to prove they can run ML and generative AI in production.
AWS describes the target candidate as having:
- At least 1 year using Amazon SageMaker AI, Amazon Bedrock and other AWS services for ML engineering
- At least 1 year in a related role such as backend software developer, DevOps developer, data engineer or data scientist
- Experience with both traditional ML and generative AI
AWS also recommends:
- Understanding of common ML algorithms, and of FM capabilities and limits
- Data engineering fundamentals, and querying and transforming data
- Software engineering practices, CI/CD and infrastructure as code (IaC)
- Knowledge of SageMaker AI, Amazon Bedrock, AWS data services, monitoring tools and AWS security best practices
There are no required prerequisites. If you are new to AI or AWS, a foundational exam first is the sensible order. I cover who MLA-C02 suits, and who should wait, in is the AWS Machine Learning Engineer – Associate worth it.
How to prepare: a step-by-step plan
Start with a diagnostic by domain, fix the heaviest weak areas first, build one small project across SageMaker AI and Bedrock, then book on evidence.
- Read the official exam guide end to end. Each domain page lists tasks and skills. Use it as a checklist and mark each skill as strong, shaky or unknown.
- Read the MLA-C01 vs MLA-C02 comparison. Anything on the "added" list needs fresh study. Anything on the "removed" list can drop down your priorities.
- Take a diagnostic before any course. Find out which of the four domains are weak. Time spent on what you already know is time taken from a gap.
- Prioritise by weight × weakness. Domain 1 carries the most weight at 28%. But scoring is overall and the other three are equal, so do not leave any domain badly weak.
- Build one small project that touches every domain. For example: prepare a dataset with AWS Glue, train and register a model in SageMaker AI, deploy it to an endpoint, then add a Bedrock knowledge base over your own documents with Guardrails, and watch both in CloudWatch. Scenario questions are much easier once you have hit the problems for real.
- Learn the trade-offs, not just the services. Know when a managed AI service beats a custom model, when RAG beats fine-tuning, when batch beats real-time, and when provisioned beats on-demand.
- Practise with scenario questions and review every wrong answer. Write down the domain and the detail you missed. Turn each mistake into a flashcard and revisit it until it sticks.
- Book when every domain is consistently strong in practice. My rule of thumb is around 80% per domain, held over several sessions. That is a personal heuristic, not an AWS figure and not a guarantee.
Prepare with StudyTech AI
StudyTech AI is a personalised AWS and AI certification prep platform. It finds your knowledge gaps, builds your practice around your skills and gaps, turns your mistakes into flashcards, scores your readiness for every exam domain, and tells you when you are ready to take your exam.
For MLA-C02, that means you start with a diagnostic across all four domains and spend your time on the weakest ones first.
Pricing is $29 a month or $297 a year, and the diagnostic is free.
- Try 15 free MLA-C02 practice questions, mapped to the four domains with full explanations
- See the MLA-C02 certification page for the full exam overview
Start a free MLA-C02 diagnostic on StudyTech AI
Cite this page
StudyTech AI. "What Is on the AWS Machine Learning Engineer – Associate (MLA-C02) Exam? Domains, Format and How to Prepare (2026)." studytech.ai/blog/aws-machine-learning-engineer-associate-exam-guide. Updated 8 November 2026.
Related reading
- AI certifications on StudyTech AI
- AWS practice exams on StudyTech AI
- Is the AWS Machine Learning Engineer – Associate worth it in 2026?
- What is on the AWS Generative AI Developer – Professional (AIP-C01) exam?
Verified 8 November 2026 against the official AWS Certified Machine Learning Engineer – Associate (MLA-C02) exam guide, its four domain pages, the AWS MLA-C01 vs MLA-C02 comparison, the AWS certification page and the AWS beta exam policy. MLA-C02 is in beta and AWS can change the format at general availability, so confirm the current details before you book.