MLA-C02 Exam Guide: What’s Changing From MLA-C01

AWS is replacing the Machine Learning Engineer – Associate exam. The last day to take MLA-C01 in English is September 28, 2026. The MLA-C02 beta starts September 29, 2026 at a $75 beta price, with 85 questions in 170 minutes. AWS has scheduled the standard version of MLA-C02 for January 14, 2027.

The short version: the four domains stay the same, but MLA-C02 adds a lot of generative AI. Think Amazon Bedrock, retrieval augmented generation (RAG), agents, guardrails, and ways to grade LLM output. This guide gives you the side-by-side numbers, exactly what AWS added and cut, and a simple way to decide whether to sit MLA-C01 while you still can or wait for MLA-C02.

MLA-C02 key dates

AWS published this timeline in its September 2026 certification update. These are the dates as announced in September 2026.

• September 1, 2026: MLA-C02 beta registration opened (English only) and the new exam guide went live.
• September 28, 2026: last day to take MLA-C01 in English.
• September 29, 2026: MLA-C02 beta delivery begins.
• January 14, 2027: MLA-C02 becomes generally available in all languages, and MLA-C01 retires in all languages.

If you test in Japanese, Korean, or Simplified Chinese, MLA-C01 stays open to you until January 14, 2027. Only the English version ends in September.

MLA-C01 vs MLA-C02 side by side

Two cells need a closer look. First, 85 questions and 170 minutes describe the beta. AWS’s beta policy says a beta exam has more total questions and more scored questions than the standard version, so don’t assume the January exam will run 170 minutes. Second, $75 is beta pricing. Check the official exam page for the standard version’s details when it launches.

One small thing if you register: AWS’s exam page lists the beta under the code ME1-C02, so look for that code when you book.

Same four domains, slightly different weights

AWS didn’t add or remove a domain. Every MLA-C01 task statement maps to an MLA-C02 task with the same number, rewritten to say “ML and AI” instead of just “ML.” Two weights move by two points.

The domain names tell the story. Domain 2 becomes “ML Model and Foundation Model (FM) Development.” Domain 4 becomes “Operating, Monitoring, and Securing ML and AI Solutions.” Deployment picks up weight, which makes sense: agents, knowledge bases, and RAG pipelines all have to be deployed and kept running somewhere.

What’s new in MLA-C02

AWS’s MLA-C01 vs MLA-C02 comparison page lists every added skill. Here they are grouped by domain, in plain English.

Data preparation (Domain 1)

• Setting up vector databases for AI apps, such as Amazon OpenSearch Service, Amazon RDS with pgvector, or Amazon S3.
• Ingesting text, images, and audio, and using embedding models to turn text and images into vectors.
• Preparing documents for RAG: chunking strategies and metadata extraction.
• Masking, redacting, and anonymizing data.
• Preparing data for foundation model fine-tuning, continued pre-training, and distillation.
• Validating AI training data, including prompt-response pairs and content safety screening.

Model development (Domain 2)

• Choosing foundation models in Amazon Bedrock based on the task and performance needs.
• Picking fine-tuning strategies and RAG architecture patterns.
• Weighing custom models against managed services, pre-trained models, and FMs, including the tradeoffs between performance, latency, and cost.
• Customizing with prompt engineering and fine-tuning, and tuning retrieval and embedding models.
• Evaluating AI output: human-in-the-loop review, NLP metrics such as BLEU, ROUGE, and BERTScore, and LLM-as-a-judge frameworks.

Deployment and orchestration (Domain 3)

• Choosing and configuring FM deployment and hosting options.
• Bringing models built outside AWS into SageMaker AI or Amazon Bedrock Custom Model Import.
• Deploying agents with their tool integrations and communication protocols, and managing agent state.
• Building Amazon Bedrock knowledge bases and retrieval pipelines, including retrieval strategies and reranking.
• Scaling GPU workloads, and automating prompt management, prompt testing, and version pipelines for agents and fine-tuned models.

Operations, cost, and security (Domain 4)

• Monitoring agents for coordination failures, truncated streaming, and tool failures.
• Monitoring FM performance, for example with Amazon Bedrock evaluations.
• Managing inference costs: token usage, embedding compute, and vector storage.
• Scanning CI/CD pipelines for code and image vulnerabilities with Amazon CodeGuru or Amazon Inspector.
• Choosing the right credential to reach FMs (Amazon Bedrock API keys vs IAM credentials).
• Adding guardrails and sensitive-data protection, such as Amazon Bedrock Guardrails.

The in-scope services list changed too. MLA-C02 adds Amazon Bedrock AgentCore and Amazon Inspector, and it drops several services the MLA-C01 list included, such as Amazon Kendra, Amazon Q, Amazon Fraud Detector, and the Amazon Lookout services. If you’ve never opened the Bedrock console, that’s your biggest gap.

What MLA-C02 drops

The removals are short. AWS took these skills out of the task statements:

• Configuring data to load into training resources such as Amazon EFS and Amazon FSx.
• Fine-tuning pre-trained models with custom datasets in Bedrock or SageMaker JumpStart, as MLA-C01 worded it. Fine-tuning now shows up inside the new FM skills instead.
• Shrinking models by changing data types, pruning, or compression.
• Optimizing models for edge devices with SageMaker Neo.
• Bring your own container (BYOC) with SageMaker.
• Monitoring infrastructure with Amazon EventBridge events.
• Troubleshooting capacity problems that involve cost and performance, such as provisioned concurrency and service quotas.

Don’t over-read this list. EventBridge, EFS, and FSx are still on the MLA-C02 in-scope services list, so they can still show up inside a scenario. You just need less depth on those specific tasks.

Should you take MLA-C01 before it ends or wait for MLA-C02?

AWS’s own advice for English test-takers is short: take MLA-C01 by September 28 if you’re already prepared and want the certification now, or take the beta if you want your certification to cover both traditional ML and generative AI. Here’s how to apply that to your situation.

Take MLA-C01 by September 28 if…

• You’re already scoring well on timed practice sets and could sit the exam within days.
• You can still find a Pearson VUE slot, at a test center or online, on or before September 28, 2026.
• You want the credential for a job application or review cycle soon.

Know the catch: this is one shot. AWS makes you wait 14 calendar days after a failed attempt, so a fail in late September puts your retake date after English MLA-C01 is gone. Your fallback would be MLA-C02, with its extra generative AI content. A pass earns the same AWS Certified Machine Learning Engineer – Associate credential, valid for 3 years.

Take the MLA-C02 beta if…

• You already build with Amazon Bedrock, RAG, or agents at work, so the new material is familiar.
• You want your certification to reflect generative AI skills, not only classic ML.
• You’re fine with one attempt. AWS lets you take a beta only once; if you don’t pass, you wait for the standard version to try again.
• You can stay sharp for 85 questions over 170 minutes.

The $75 price is tempting, but don’t let it make the decision. Prep material written for MLA-C02 will be thin at first, since the exam guide only came out on September 1, 2026. Also, don’t sit on the decision for months: AWS says beta exams generally run for 1 to 3 months. Results come on the normal timeline, within 5 business days, and a beta pass earns the full certification.

Wait for the standard MLA-C02 if…

• You’re starting from scratch or have little hands-on Bedrock experience.
• You want normal retake rules (14-day wait, no attempt limit) instead of a single beta attempt.
• You’d rather study from materials built around the final exam.

AWS has scheduled general availability for January 14, 2027. Use the months in between to get hands-on with Bedrock knowledge bases, agents, and guardrails. If you’re still weighing whether the credential is worth the effort at all, read our honest take on the AWS ML Engineer Associate.

Testing in Japanese, Korean, or Simplified Chinese

You have more runway. MLA-C01 stays available in those languages until January 14, 2027, when MLA-C02 launches in all languages and MLA-C01 retires everywhere.

What this means for your study materials

Your MLA-C01 prep isn’t wasted. Every C01 task statement carries over to a C02 task, and the domain weights barely move. Most of what you know about SageMaker AI, data pipelines, CI/CD, monitoring, and IAM still applies.

What changes is the layer on top. A practical way to split your time:

• Keep: data prep with AWS Glue, Amazon Athena, and Amazon S3; SageMaker AI training, tuning, and endpoints; pipelines and CI/CD; model monitoring; IAM and AWS KMS.
• Add: Bedrock model selection, knowledge bases, RAG chunking and retrieval, agents and AgentCore, guardrails, prompt management, and LLM evaluation metrics.
• Trim: SageMaker Neo, BYOC, and model compression techniques.

Our free AWS ML Engineer practice questions and the MLA-C01 study guide and cheat sheet follow the MLA-C01 blueprint. They’re the right tools if you’re sitting C01 by September 28, and they still cover the shared core if you’re heading to C02. For the new generative AI topics, work from the official MLA-C02 exam guide. AWS’s MLA-C02 page also points to an Exam Prep Plan on AWS Skill Builder and the AWS Certification Official Practice Question Set.

This prompt turns AWS’s list of additions into a personal plan. Copy the “Additions of content for MLA-C02” section from the comparison page and paste it where shown.

You are an AWS machine learning instructor helping me move from MLA-C01 prep to the MLA-C02 exam.My background: [your role, years with SageMaker, any Amazon Bedrock experience]Hours I can study per week: [number]My target exam date: [date]Below is AWS's official list of skills added in MLA-C02. Using ONLY this list:1. Group the skills into 4-6 study themes (for example: RAG and knowledge bases, agents, FM evaluation).2. For each theme, rate how new it is likely to be for someone with my background (new / partly familiar / familiar) and explain why in one sentence.3. Build a week-by-week plan that fits my hours, with one hands-on lab idea per week that I can do in the AWS console.4. End with 10 practice questions in MLA-C02 style (one correct answer, three distractors), with answers and one-line explanations.Do not invent AWS service features. If you are not sure a service can do something, say so instead of guessing.[PASTE THE "ADDITIONS OF CONTENT FOR MLA-C02" SECTION HERE]

You are an AWS machine learning instructor helping me move from MLA-C01 prep to the MLA-C02 exam.My background: [your role, years with SageMaker, any Amazon Bedrock experience]Hours I can study per week: [number]My target exam date: [date]Below is AWS's official list of skills added in MLA-C02. Using ONLY this list:1. Group the skills into 4-6 study themes (for example: RAG and knowledge bases, agents, FM evaluation).2. For each theme, rate how new it is likely to be for someone with my background (new / partly familiar / familiar) and explain why in one sentence.3. Build a week-by-week plan that fits my hours, with one hands-on lab idea per week that I can do in the AWS console.4. End with 10 practice questions in MLA-C02 style (one correct answer, three distractors), with answers and one-line explanations.Do not invent AWS service features. If you are not sure a service can do something, say so instead of guessing.[PASTE THE "ADDITIONS OF CONTENT FOR MLA-C02" SECTION HERE]

Check any AI-written practice question against AWS documentation before you trust it. Chatbots can describe Bedrock features that don’t exist with total confidence, so run a quick 3-minute fact check on anything that surprises you.

If you already hold MLA-C01

Nothing changes for you today. AWS says your credential stays active through its original expiration date.

When renewal comes up, AWS’s recertification page lists three options for the ML Engineer – Associate: pass the latest version of the exam (MLA-C02 by then), pass the AWS Certified Generative AI Developer – Professional exam, or maintain the certification on AWS Skill Builder with a paid subscription, which extends it one year at a time. You can use the 50% discount voucher in your AWS Certification Account for the exam routes.

Want to sit MLA-C02 anyway to prove the newer skills? AWS normally won’t let you retake an exam you’ve passed for two years, but you’re eligible when the exam gets a new exam guide and series code, which MLA-C02 has. Bonus: passing the latest ML Engineer – Associate exam also renews an AWS Certified AI Practitioner credential.

Common mistakes to avoid

• Treating the beta as a cheap practice run. You get one attempt. A fail means waiting for the standard exam.
• Booking MLA-C01 on the last possible day. AWS lets you reschedule an appointment only twice, and after September 28 there’s no English MLA-C01 date to move to.
• Assuming the standard MLA-C02 will be 170 minutes. Beta exams are longer by design. Check the exam page after January 14, 2027.
• Reading about Bedrock instead of using it. The new skills are about configuring knowledge bases, agents, and guardrails. Clicking through each one once in the console beats rereading a summary.
• Skipping the removed-topics list. Every hour on SageMaker Neo or BYOC is an hour you didn’t spend on RAG.

Try it now: pick your path in 15 minutes

1. Take a timed set of our free MLA practice questions and note your weakest domain.
2. If it’s still before September 28, 2026 and you’re scoring well, book English MLA-C01 now.
3. Otherwise, open AWS’s comparison page, run the prompt above, and see how big your generative AI gap really is.
4. Decide: the one-attempt beta, or the standard MLA-C02 from January 14, 2027. More free practice sets for other AI certs live on our AI exam prep hub.

Related guides

• AWS ML Engineer Associate (MLA-C01) Study Guide & Cheat Sheet
• Free AWS Machine Learning Engineer – Associate Practice Questions
• MLS-C01 Is Retired: What Replaced AWS’s ML Specialty (and What to Take Now)

HOW TO // AI is not affiliated with or endorsed by Amazon Web Services. Exam details come from AWS’s official certification pages, exam guides, and policy pages as of September 2026 and can change. Check the official exam page before you book.