Get Your CertGuidesPlayCompete
‹ All study guides

AWS ML Engineer Associate, MLA-C01 to MLA-C02

Updated 14 October 2026 · 7 min read

The AWS Certified Machine Learning Engineer – Associate exam has a new version. The last day to sit MLA-C01 in English was September 28, 2026, and the beta of MLA-C02 has been delivered since September 29, 2026. If you were halfway through MLA-C01 study, most of it still counts. What you need to add is a solid layer of generative AI engineering: foundation models on Amazon Bedrock, retrieval-augmented generation (RAG), agents, and the evaluation, cost and safety work that comes with them.

This guide sets out what AWS has published about the change, and what it hasn't. Every fact here was checked against AWS's own certification page, exam guide and Training and Certification blog on October 14, 2026.

The dates and the beta format

AWS announced the update on its Training and Certification blog in July 2026. The timeline it gives:

  • Beta registration opened on September 1, 2026, in English only, and the MLA-C02 exam guide was released the same day.
  • September 28, 2026 was the last day to take MLA-C01 in English.
  • Beta delivery began on September 29, 2026.
  • MLA-C01 stays available in Japanese, Korean and Simplified Chinese until MLA-C02 reaches general availability, which the blog describes only as "early 2027". No exact date is published.

The certification page compares the two versions side by side:

  • MLA-C01: 65 questions, multiple choice and multiple response, 130 minutes, $150 USD.
  • MLA-C02 beta: 85 questions, multiple choice and multiple response, 170 minutes, $75 USD, English only, at a Pearson VUE test center or online proctored.

The beta's extra length comes from additional unscored items. The certification page says the exam time already accounts for them and that they do not affect your score. The MLA-C02 exam guide, which describes the standard exam, lists the same two question types as before and says it has 50 questions that affect your score plus 15 unscored ones that are not identified. AWS has not published the time limit or price of the standard MLA-C02 exam that follows the beta, so don't assume either until the certification page says so.

What "beta" means for you

AWS uses beta exams to test new questions before they go into the standard version. According to AWS's exam policies and the certification page:

  • Results are not instant. AWS says beta results are typically available within five business days of finishing the exam, in your AWS Certification account.
  • Passing a beta earns the same certification as the standard exam, active for three years.
  • Beta exams are scored against a minimum standard set by AWS. The exam guide gives a scaled score of 100 to 1,000 with 720 to pass, but marks its results section as not applying to the beta, so the beta's own passing standard is not published as a number.
  • The beta is discounted: $75 against $150 for MLA-C01.

The practical trade-off is cheaper and earlier, against a longer sitting, no immediate result, English only, and less third-party prep material written for the new content. If you need the credential this quarter and you are comfortable with Bedrock, the beta is a reasonable bet. If you are still learning the GenAI parts from scratch, waiting for general availability costs you nothing but time.

Domains and weights

AWS says plainly that the domain structure is the same and no new domains were added. Each of the four domains has been renamed to take in AI as well as ML, and two weights moved by two points:

  • Domain 1, data preparation: "Data Preparation for Machine Learning (ML)" becomes "Data Preparation for ML and AI". 28% in both versions.
  • Domain 2, development: "ML Model Development" becomes "ML Model and Foundation Model (FM) Development". Down from 26% to 24%.
  • Domain 3, deployment: "Deployment and Orchestration of ML Workflows" becomes "Deployment and Orchestration of ML and AI Workflows". Up from 22% to 24%.
  • Domain 4, operations: "ML Solution Monitoring, Maintenance, and Security" becomes "Operating, Monitoring, and Securing ML and AI Solutions". 24% in both versions.

Every MLA-C01 task statement maps to the task with the same number in MLA-C02, so the skeleton you studied is intact. The change is in the skills listed under each task. The exam guide also states a new target candidate: at least a year of experience with SageMaker AI, Amazon Bedrock and other AWS services for ML engineering, and experience with both traditional ML and generative AI.

What was added

The exam guide has a page comparing the two versions. The additions cluster into a few themes:

  • Data for GenAI. Configuring vector stores (OpenSearch Service, Amazon RDS with pgvector, Amazon S3), using embedding models for text and images, tokenization, chunking and metadata extraction for RAG, masking and redacting data, and preparing data for fine-tuning, continued pre-training and distillation. Checking prompt-and-response training pairs and screening content for safety are also new.
  • Choosing and customizing models. Selecting a foundation model on Bedrock, choosing a fine-tuning strategy, weighing custom solutions against managed services and foundation models, picking a RAG pattern, and trading off quality, latency and cost.
  • Evaluating GenAI. Human evaluation workflows, NLP metrics such as BLEU, ROUGE, BERTScore and semantic similarity, LLM-as-a-judge, bias detection in outputs, and measuring retrieval accuracy in a RAG system.
  • Deploying GenAI and agents. Foundation model deployment options, bringing outside models in with Bedrock Custom Model Import, creating and tuning Bedrock knowledge bases, retrieval strategies and reranking, deploying agents and their state management, GPU scaling, prompt management with Bedrock Prompt Management, prompt testing, agent version management, and pipelines that refresh a knowledge base.
  • Operating GenAI. Monitoring agents for tool failures and coordination problems, Bedrock evaluations in production, CloudWatch generative AI observability, Bedrock AgentCore Observability, token and embedding cost tracking, and vector storage costs.
  • Security. Scanning CI/CD pipelines for code and image vulnerabilities, choosing between Bedrock API keys and IAM credentials, and applying Amazon Bedrock Guardrails for sensitive data and responsible AI policies.

The in-scope services list now names Amazon Bedrock AgentCore alongside Amazon Bedrock and SageMaker AI.

What was removed

The comparison page lists only a handful of deletions:

  • Configuring data to load into training resources such as Amazon EFS and Amazon FSx.
  • Using custom datasets to fine-tune pre-trained models with Bedrock or SageMaker JumpStart, now covered by the broader fine-tuning and customization skills.
  • Reducing model size by pruning, compression or changing data types.
  • Optimizing models for edge devices with SageMaker Neo.
  • Bring your own container (BYOC) with SageMaker.
  • Monitoring infrastructure with EventBridge events, and troubleshooting capacity issues involving provisioned concurrency, service quotas and auto scaling, from the cost and performance task.

Removed from the skills list is not the same as banned from the exam. EventBridge, EFS and FSx are all still on the in-scope services list, so keep the basics.

Closed SageMaker features

Several SageMaker AI features that MLA-C01 study material leans on are no longer open to new customers, according to banners on their documentation pages: SageMaker Clarify, Model Monitor, Ground Truth, Debugger and Augmented AI among them. Existing customers keep using them, but AWS plans no new features. The MLA-C02 skills don't name any of these. They still ask you to detect drift, measure bias, label data, explain outputs and debug convergence, so learn the concepts and the metrics, and treat the tools as something you may still see in an older account rather than what you would build with today. Skill 1.2.1 still names SageMaker Data Wrangler, which AWS has integrated into SageMaker Canvas.

How to prepare if you studied MLA-C01

Your MLA-C01 work carries over in four big blocks:

  1. Data engineering. Ingestion with Kinesis and Firehose, Glue jobs and Glue Data Quality, formats such as Parquet, Feature Store, splits, leakage and imbalance. Domain 1 still carries the largest weight.
  2. Training and tuning. Built-in algorithms, script mode, automatic model tuning, distributed training, overfitting, and metrics for imbalanced classes.
  3. Deployment and MLOps. Real-time, serverless, asynchronous and batch inference, shadow tests, auto scaling, SageMaker Pipelines, Model Registry and managed MLflow.
  4. Security and cost. IAM least privilege, VPC endpoints, KMS, CloudTrail and Config, instance and purchasing choices.

Then add the GenAI layer in this order, since each part builds on the one before:

  1. Bedrock basics: choosing a model, inference parameters, prompt design, and the difference between on-demand use and the other hosting options.
  2. RAG end to end: chunking, embeddings, vector stores, Bedrock Knowledge Bases, syncing, and reranking. Our guide on prompt engineering, RAG and fine-tuning covers when each lever fits.
  3. Customization: preparing fine-tuning and distillation data, and versioning custom models.
  4. Evaluation: automatic metrics, human review and LLM-as-a-judge, for both model output and retrieval quality.
  5. Agents: AgentCore deployment, state, observability, and versioning agents in a pipeline.
  6. Guardrails, Bedrock API keys versus IAM credentials, and token-level cost monitoring.

Read the four domain pages of the MLA-C02 exam guide line by line and mark each skill you could not explain to a colleague. That list is your study plan, and it is more reliable than any third-party summary, because AWS published the guide on the day beta registration opened.

Practice questions

These MLA-C01 questions test skills that are still on the MLA-C02 outline, and none of their answers depends on a closed feature:

Further reading

Practice questions

AWS Certified Machine Learning Engineer – Associate (MLA-C01)100 questions with explanations