AWS Machine Learning Certification 2026: MLA-C02 Change, Cost + Prep

Last updated: September 2026. Written by Josh Hutcheson, OnlineCourseing editor. Exam facts, dates and prices re-read directly on AWS’s certification pages and the MLA-C01 exam guide on 20 September 2026; every prep option loaded and checked live the same day. See our review methodology.

THE 60-SECOND VERSION

VERIFIED ON AWS’S OWN PAGES, 20 SEPTEMBER 2026

Start MLA-C02 prep (course already updated for C02) →

Every MLA-C02 date that matters

AWS publishes these as a “Key dates” list on the certification page. They are worth reading carefully, because the one people misread – 28 September – is narrower than it looks: it ends the English MLA-C01 exam, not the exam.

IF YOU HOLD THE CERTIFICATION ALREADY

What is actually different in MLA-C02

The short answer is generative AI. AWS says the update “reflects the broadened scope of the ML engineer role” and lists four additions:

• Generative AI implementation – Amazon Bedrock and RAG architectures.
• Agentic AI – orchestrating AI agents and multi-step workflows.
• Foundation models and LLMs – selection, fine-tuning and operationalisation.
• Responsible AI – applied across traditional ML and generative AI alike.

The structural point matters more than the list, though: AWS states the domain structure is unchanged and no new domains were added. The four domains and their weightings carry over; what changed is the task statements inside them. So a study plan built on the MLA-C01 domain map is still the right shape – you are adding Bedrock, RAG and agent orchestration to each domain rather than learning a new blueprint.

You can see the same shift in the intended-candidate line. MLA-C01 asks for about a year with “Amazon SageMaker and other ML engineering AWS services”. MLA-C02 asks for a year with “Amazon SageMaker AI, Amazon Bedrock, and other ML engineering AWS services”, and adds LLMOps engineer to the list of example job roles. Bedrock has moved from adjacent to assumed.

THIS IS AN INDUSTRY MOVE, NOT AN AWS ONE

MLA-C01 or MLA-C02: which should you sit?

This is the only real decision on this page, and it turns on how ready you are today rather than on any judgement about the exams. Find your situation:

One thing not to weigh: prestige. Both routes award the identical credential, AWS Certified Machine Learning Engineer – Associate, valid three years. A verifier cannot tell which version you sat, and no employer asks.

See the MLA-C02-updated course on Udemy →

Exam format, side by side

AWS lists the MLA-C02 exam code as ME1-C02 on its comparison table while calling the exam MLA-C02 everywhere else; the Udemy course that has updated for it uses “MLA-C02 / ME1-C02” too. Book by the exam name rather than the code and you will not go wrong.

The four domains and their weightings

From the MLA-C01 exam guide, and unchanged in structure for C02. The percentages are of scored content:

Notice how back-loaded this is. Domains 3 and 4 together are 46% of the exam and both are about running a model in production – endpoints, pipelines, monitoring, access control. Only 26% is model development. That is the difference between this exam and the retired Specialty, and it is why candidates who come from notebooks rather than deployment find it harder than the Associate label suggests.

The exam guide is also blunt about what is out of scope, which is useful for not over-studying: architecting full end-to-end ML solutions, setting ML strategy, deep specialisation in two or more ML domains such as NLP plus computer vision, and quantising models to analyse accuracy impact.

How it is scored (and how to use that)

• 720 out of 100–1,000, scaled. Scaled scoring equates results across exam forms of slightly different difficulty.
• Compensatory. You need to pass the exam overall, not each domain. AWS says so explicitly.
• 50 of the 65 questions count on MLA-C01. The other 15 are unscored items AWS is evaluating, and they are not flagged.
• No penalty for guessing – unanswered questions score as incorrect, so answer everything.
• More than multiple choice. The guide lists ordering, matching and case-study formats alongside multiple choice and multiple response. Practice tests that only drill four-option questions under-prepare you for these.

THE PRACTICAL CONSEQUENCE

Best prep for MLA-C02, honestly ranked

We checked every option below live on 20 September 2026. The transition changes the ranking: most MLA prep was written for C01, and Bedrock, RAG and agentic AI are now examinable. Course currency is the deciding factor this quarter, not production values.

1. Udemy – AWS Certified Machine Learning Engineer Associate: Hands On! (Frank Kane + Stephane Maarek)

The pick, on one specific and checkable merit: it is the only prep we found that is already updated for MLA-C02. Its own description reads “FULLY UPDATED for the MLA-C02 exam released in September 2026”, and it covers SageMaker, Bedrock and the AI material the new version adds. Ratings hold up too – 4.5 from 6,028 ratings, 62,307 students, last updated September 2026. A practice exam is included. Usually $15–$30 in a Udemy sale.

We do not normally lead an Associate-level cloud guide with Udemy, and we are here for a narrow reason: currency beats depth in a syllabus-change quarter. If you sit MLA-C02, this is the only course we verified that teaches the exam you are actually taking.

Take the MLA-C02-updated course →

2. AWS Skill Builder Exam Prep Plan – free, official, and the C02 source of truth

AWS points MLA-C02 candidates at its own four-step Exam Prep Plan, and for a brand-new exam version the vendor gets there first: the official practice question set, the official pretest and the official practice exam are updated before third parties are. It is free to browse and there is no affiliate programme behind it, so we get nothing from this recommendation – take it anyway, and pair it with a course for the explanation Skill Builder does not give you. One detail worth reading there: the exam now uses short names for some AWS services, with a full-name lookup behind the Help button.

3. Udacity AWS Machine Learning Engineer Nanodegree (nd189) – the hands-on route, with a caveat

Still the deepest build-it-yourself option: 94 hours at intermediate level, shipping an image classifier to a SageMaker endpoint and wiring workflows with Lambda and Step Functions. That is domains 3 and 4 – 46% of the exam – done with your hands rather than read about.

The caveat is the reason it is not first. Its curriculum is MLA-C01-shaped: SageMaker, Lambda, Step Functions, distributed training. We found no Bedrock, RAG or agentic-AI content in it. For the skills, it is excellent and still our recommendation for anyone who wants portfolio projects. For MLA-C02 coverage, it is incomplete, so treat it as the engineering half and get the generative-AI half elsewhere.

On price, read the page carefully: Udacity shows $249/month list, discounted to $150/month, a four-month prepay at $507.60, and a single-program option at $999 reduced to $599.40. For a program most people finish in five to six months, the one-off is usually the cheaper door. Our link carries a 40% code, which is the same 40% Udacity is currently showing publicly – it is attribution for us, not an extra discount for you, and we would rather say so. Read our full review of the Nanodegree.

Check the current Udacity price →

4. Coursera – Exam Prep MLA-C01 specialization

A five-course exam-prep series on Coursera, taught by a Whizlabs instructor, structured around roughly two hours a week at intermediate level. It is live and it is the most guided path of the paid options. Two honest marks against it: it is titled for MLA-C01, so expect the same generative-AI gap as the Nanodegree until it is revised, and it carries only twelve reviews, which is thin evidence either way. Reasonable if you already hold Coursera Plus; otherwise the Udemy course is better value and more current.

See the Coursera exam-prep series →

5. Practice tests

Tutorials Dojo (Jon Bonso) remains the community default for AWS practice tests and is worth the roughly $15 – we have no affiliate relationship with them, so that is a straight recommendation. AWS’s own official practice exam is the other one to buy, and for MLA-C02 specifically it will be accurate before anyone else’s is. Whatever you use, check the listing says C02 before paying: Udemy already has practice sets labelled for both versions, and a C01 set will drill you on a syllabus with no Bedrock in it.

ONE COURSE WE CHECKED AND DID NOT PICK

What it costs, all in

A lean, realistic budget for MLA-C02 is about $110 to $190: the $75 beta, one course, one practice set, and a small AWS bill. That is materially cheaper than the same path was a month ago, purely because the beta is half price. If you want the guided route, add the Nanodegree or the Coursera series and the number moves into the several hundreds.

THE COST NOBODY BUDGETS

A ten-week plan that fits the transition

Built for MLA-C02, since that is the right sit for most people reading this after 28 September. Ten weeks at roughly ten hours, front-loading a diagnostic so compensatory scoring works for you:

1. Week 1 – Read the exam guide end to end, then take a full practice exam cold. You will fail it; the point is the domain breakdown.
2. Weeks 2–3 – Domain 1. Ingestion and storage, Glue transforms, Feature Store, data quality and validation. Build one real pipeline.
3. Weeks 4–5 – Domain 2. Built-in algorithms, training jobs, hyperparameter tuning, AutoML, evaluating and versioning models.
4. Week 6 – The generative-AI additions. Bedrock, a RAG implementation, foundation-model selection and fine-tuning, an agent workflow. This week is new to C02 and skipped by C01 courses.
5. Weeks 7–8 – Domain 3. Deploy the same model four ways – real-time, asynchronous, serverless, batch transform – then orchestrate it with Step Functions and put it behind CI/CD.
6. Week 9 – Domain 4. Model Monitor, CloudWatch, drift detection, IAM scoping for ML, encryption, cost controls. Plus responsible-AI practice across both ML and GenAI.
7. Week 10 – Your two weakest domains from week 1, re-tested. Then the official practice exam, and book.

If you are sitting MLA-C01 in the next few days instead, compress to the diagnostic, domains 3 and 4, and two practice exams – and skip week 6 entirely, because Bedrock and agents are not on C01.

WEEK 6 IS THE ONE TO CHECK BEFORE YOU BUY

See a curriculum that covers week 6 →

What this certification actually signals

We used to publish salary bands here. We have removed them: they were not sourced to anything we could verify, AWS does not publish compensation data, and the aggregators that do block automated checks – so we were repeating numbers we could not stand behind. What we can tell you is what the exam demonstrably tests, which is the part an interviewer probes.

Per AWS’s own exam guide, a holder has shown they can ingest and prepare data for modelling, train and tune models, choose deployment infrastructure and endpoint types, provision compute and configure auto scaling, set up CI/CD to orchestrate ML workflows, monitor models and infrastructure for problems, and secure ML systems with access controls and compliance features. With MLA-C02 that list extends to standing up Bedrock and RAG systems and operating agentic workflows responsibly.

In hiring terms that is the operational half of ML, which is where teams that already have working notebooks get stuck. AWS’s own role examples for MLA-C02 – ML engineer, MLOps engineer, LLMOps engineer, data engineer, backend developer, data scientist – are a fair map of who the credential reads well to. It is not a research-scientist credential and does not pretend to be. If you want the broader picture of which credentials employers ask for, we compare the field in best AI certifications and best cloud certifications.

Where it sits among AWS’s AI and data certifications

The common pairing is Data Engineer Associate plus ML Engineer Associate, because together they cover the whole path from raw data to a monitored endpoint. If you want the credential without the engineering, the AI Practitioner is the honest answer. And if you are weighing clouds rather than exams, our AWS vs Azure comparison covers how the two certification ladders differ.

Frequently asked questions

Is MLA-C01 being retired?

It is being replaced, in stages. AWS’s certification page gives 28 September 2026 as the last day to take MLA-C01 in English, and says the exam remains available in Korean, Japanese and Simplified Chinese until MLA-C02 reaches general availability. So MLA-C01 is not switched off globally on 28 September – only the English version is. Either way, a pass you already hold stays valid for its full three years; AWS states the certification remains active through its full validity period regardless of when you earn it.

Should I take MLA-C01 or MLA-C02?

If you are exam-ready right now and can sit before 28 September 2026, MLA-C01 is the safer sit: its exam guide, courses and practice tests are mature. If you are more than a week or two from ready, go straight to MLA-C02. It is the same credential – AWS Certified Machine Learning Engineer Associate – and the beta is $75 instead of $150, so waiting is cheaper as well as more current.

How much does the MLA-C02 beta cost?

$75 USD, which AWS lists as beta pricing, against $150 USD for MLA-C01. Beta registration opened on 1 September 2026 and beta delivery begins 29 September 2026. The trade-offs are that the beta runs longer (170 minutes, 85 questions rather than 130 minutes and 65), it is English only, and results take about five business days instead of arriving at the test centre.

Why does the MLA-C02 beta have more questions?

Because beta exams carry extra unscored items that AWS is trialling for future use. AWS says the additional time and questions account for those items and that they do not affect your score. It is the same reason MLA-C01 has 65 questions of which only 50 are scored.

What is actually new in MLA-C02?

Generative AI, mostly. AWS lists four additions: generative AI implementation with Amazon Bedrock and RAG architectures, agentic AI covering orchestration of AI agents and complex workflows, foundation models and LLMs including selection, fine-tuning and operationalisation, and responsible AI practices across both traditional ML and generative AI. AWS is explicit that the domain structure is unchanged and no new domains were added – the existing task statements were rewritten.

What happened to the AWS Machine Learning Specialty (MLS-C01)?

It was retired. The last day to take MLS-C01 was 31 March 2026, and AWS says existing holders keep an active certification for three years from the date they earned it. There is no route to earning it now. The ML Engineer Associate is the replacement, and it is deliberately a different exam: Associate level rather than Specialty, and weighted towards putting models into production rather than choosing algorithms.

What score do I need to pass?

720 on a scaled range of 100 to 1,000, per the MLA-C01 exam guide. Scoring is compensatory, which means you do not have to pass each domain individually – only the exam overall. That matters for planning: a weak domain can be carried by strong ones, so the efficient strategy is lifting your worst two domains to competent rather than perfecting your best.

How long should I study for it?

Eight to twelve weeks at around ten hours a week is realistic if you already have roughly a year of hands-on AWS and some ML exposure, which is what AWS recommends for the target candidate. If SageMaker is new to you, plan closer to sixteen weeks – the exam asks about endpoint types, pipeline orchestration and monitoring in operational detail that is hard to absorb without building something.

Is the ML Engineer Associate worth it if I already hold the AI Practitioner?

Usually yes, because they test different things. The AI Practitioner is a foundational, largely conceptual credential; the ML Engineer Associate is an engineering exam about deploying and operating ML systems. With MLA-C02 folding in Bedrock, RAG and agentic workflows, the two now overlap more at the vocabulary level, but the Associate still expects you to build and monitor the thing rather than describe it.

Final verdict

The AWS Certified Machine Learning Engineer – Associate is the right ML exam on AWS, and for the next few days it is also the rare case where waiting is the better move. If you can sit MLA-C01 before 28 September 2026, do it and be done. Everyone else should take the MLA-C02 beta at $75, which is half price and tests the syllabus your job actually has – Bedrock, RAG and agents included. Pair the Kane and Maarek course, which is already updated for C02, with AWS’s free Exam Prep Plan and one practice set, and budget somewhere around $110 to $190 all in. The thing that passes this exam is unchanged by the version number: hours spent deploying and monitoring something real in an account you pay for.

Start MLA-C02 prep →

Related guides: AWS AI Practitioner (AIF-C01) · AWS Data Engineer (DEA-C01) · AWS Solutions Architect Associate · AWS Developer Associate · AWS DevOps Engineer Professional · AWS Cloud Practitioner · Best cloud certifications · Best AI certifications · AWS vs Azure · Udacity AWS ML Engineer Nanodegree review · Best generative AI courses