AWS Certification Changes: What to Book Before September

If you are studying for the AWS Machine Learning Engineer – Associate exam, note this date: 28 September 2026 is the last day to sit MLA-C01 in English. Registration for the updated version opens on 1 September, and beta delivery begins on 29 September. Two other things have already changed that still catch people out — SysOps Administrator is now CloudOps Engineer, and the Machine Learning – Specialty exam has been retired entirely. If you already hold a certificate, no action is needed: it stays valid to its own expiry date and does not rename itself. Here is what changed, and what to book now.

The exam that stopped existing

A customer's son came into Jake's shop wanting a laptop that could handle "machine learning work." He'd been studying four months, he said, for the AWS Machine Learning Specialty.

"Four months," Jake told Ethan afterwards. "And I had to be the one to tell him. That exam doesn't exist any more."

"Where did he get the material?"

"A course. Still on sale. Still says Machine Learning Specialty in the title."

"Then the course took his money and nobody updated it. That happens more than people think — a certification retires, and the training material carries on selling for years because nothing forces it off the shelf."

"So how does anyone know?"

"They check the exam code. Not the course title, not the marketing — the code. Ten seconds, and it would have saved him four months."

Which is worth knowing now, because another AWS exam has a deadline seven weeks away, and the same thing is about to happen to a fresh set of people.

Three changes, one of them urgent

Only the first has a deadline. The other two are already history, and they matter because stale courses are still selling preparation for both of them.

MLA-C01 to MLA-C02: the one with a clock on it

The two versions side by side

What MLA-C02 adds

The update drags the exam into how machine learning work is actually done in 2026:

• Building and deploying generative AI solutions, not only traditional models
• Fine-tuning foundation models — adapting somebody else's model rather than training your own
• Retrieval-augmented generation, the standard pattern for answering from your own documents
• Agents and workflow orchestration — models that take actions, not just produce text
• Expanded Amazon Bedrock coverage
• Responsible AI across both traditional and generative workloads

No new domains were added — the existing ones were updated. That matters if you have been studying: your notes are a base to extend, not something to bin.

It also tells you something about the industry. Two years ago this exam was mostly about training your own models. It now assumes you will more often be adapting somebody else's — which is the same shift we walk through in Day 1 of the AI series, where foundation models are the fourth of five nested ideas.

So do you rush, or wait?

The question everyone actually has. Ethan's answer, and he is not hedging it:

If you are nearly ready, book MLA-C01 before 28 September. Your preparation matches that version, the practice material for it is mature, and you get your result quickly. Sitting the exam you studied for is worth a great deal.

If you are starting from scratch today, do not rush. Seven weeks of panicked study aimed at a retiring exam is the worst of both worlds — you risk failing an exam that is about to disappear, and every hour spent on a syllabus with a shelf life is an hour you could have spent on the one that replaces it.

If you are somewhere in the middle — a few weeks in, not confident — the honest answer is that $75 for the beta is half the price of $150 for a rushed attempt, and the additional material is the direction the whole field is moving anyway. You will wait longer for a result. You will study with thinner resources. Those are real costs and only you know whether a slower answer matters for your situation.

"And if someone needs the certificate by a date?" Jake asked. "For a job, say."

"Then the beta is the wrong choice, full stop. You cannot promise an employer a result on a timetable AWS has not committed to."

SysOps Administrator is now CloudOps Engineer

This one has already happened, and it still confuses people every week.

AWS Certified SysOps Administrator – Associate was renamed and updated to AWS Certified CloudOps Engineer – Associate, exam code SOA-C03. Registration opened 9 September 2025, the last day for the old SOA-C02 was 29 September 2025, and SOA-C03 has been the only option since 30 September 2025.

The question everyone asks: does my certificate rename itself?

No. If you hold the SysOps Administrator certification, it stays valid until its own expiry date and keeps the name it was issued under. You earn the CloudOps Engineer title only by passing SOA-C03.

That is not a demotion and it is not something to fix. A credential is a record of what you passed and when. When it comes up for renewal you will sit whatever the current version is by then, which is how recertification has always worked.

What actually changed in the exam

More than a name. Containers are now in scope, roughly twenty-five further AWS services were added, and there is much greater weight on multi-account and multi-Region architectures and on automation and infrastructure as code.

Which is a fair description of how the job itself changed. "SysOps" suggested someone looking after servers. "CloudOps" suggests someone operating distributed systems they mostly do not touch by hand. The rename is the honest part.

The Machine Learning Specialty is gone

The old AWS Certified Machine Learning – Specialty has been retired, and its ground was divided in two:

• AI Practitioner (AIF-C01) — the foundational and conceptual side, for people who need to understand and discuss AI without building it. 65 questions, 90 minutes, $100, valid three years.
• ML Engineer – Associate (MLA-C01, soon C02) — the practical build-and-deploy side.

That split is genuinely sensible. One exam was trying to serve a product manager who needed the vocabulary and an engineer who needed to deploy models, and it served neither especially well.

But it is also the trap Jake's customer fell into. A retired exam does not pull its training material off the market. Courses keep selling, ranking, and taking money for a certification nobody can sit.

How to check a course before you pay for it

Ten seconds, and it is the single most valuable habit in this whole subject.

Find the exam code, not the course title. Course names are marketing; codes are facts. Then check it against this:

A second check worth doing: look at the last updated date, and be suspicious if there isn't one. Material that cannot tell you when it was written is telling you something anyway.

So which one should you actually take?

These are exams for different jobs, not rungs on a ladder. Plenty of people only ever want the AI Practitioner, and that is a complete and reasonable outcome rather than a first step towards something else.

If the cloud part is what is new to you rather than the AI part, start with what AWS actually is and whether learning it pays, which covers the free tier and the ways a free account still bills you. For the AI side, Day 2 of the series covers clustering and inference, and problem formulation is the one that saves people from building something nobody needed.

The seven weeks, if you are sitting MLA-C01

If you decided to sit the current version before the cutoff, the useful thing is not encouragement — it is a shape for the time you have. This assumes evenings and part of a weekend, which is what most people actually have.

Book the exam this week, not later

Not because slots run out, though they can tighten near a retirement date. Because an unbooked exam is an intention and a booked one is a deadline, and the single most common way people miss a cutoff is by waiting until they feel ready. You will not feel ready. Book it, then work backwards.

Pick a date in the third week of September, not the twenty-eighth. That leaves room to reschedule once if something goes wrong, and something usually does.

Weeks one and two: find out what you do not know

Read the official exam guide first — the actual PDF, not a summary of it. It lists the domains and their weightings, and it is the only document that defines what is on the exam. Everything else is somebody's interpretation of it.

Then sit a full practice exam immediately, before studying. It will be uncomfortable and that is the point: you are not measuring readiness, you are producing a map of gaps. Studying without one means spending your best hours on the domain you already enjoy.

Weeks three to five: build something small

This is the part people skip and the part that separates a pass from a scrape. Associate-level AWS exams ask scenario questions, and scenario questions are answered from having done the thing rather than having read about it.

Train a small model. Deploy it to an endpoint. Break it deliberately — wrong permissions, wrong instance type, an input the model has never seen. Then take it down again so it stops billing you.

"And if the whole point is that he has not got time?" Jake asked.

"Then he has time for one small thing rather than three big ones," Ethan said. "One deployment he did himself is worth more in that exam room than a week of watching somebody else do six."

Weeks six and seven: practice exams and the weak domain

Practice exams until you are consistently comfortable, and spend the remaining time on whichever domain keeps scoring worst. Not the one you find most interesting — the one that keeps costing you marks. Those are rarely the same domain, and that gap is where most near-misses live.

The night before, stop. Nothing learned in the final evening survives the exam room, and being tired costs more marks than one extra topic gains.

Test centre or online? The decision people rush

You choose this at booking and most people click whichever is cheaper to reach. It is worth thirty seconds of thought, because the two fail in completely different ways.

Online proctored means sitting it at home with a webcam and a proctor watching. No travel, more available slots, and you can book far closer to the date — which matters when there is a cutoff. The costs are real though: your room must be empty of notes and other people, your desk clear, and your internet stable for the full session. Connection problems mid-exam are the horror story, and they do happen.

A test centre means travelling, fewer slots, and booking further ahead. In exchange you get a machine that is not yours, a network that is not your responsibility, and nobody walking into the room. If your home internet is unreliable or your household is busy, this is the safer choice and it is not close.

Ethan's rule: if you would be anxious about your connection, that anxiety will cost you more marks than the travel would. Pay the inconvenience rather than carrying the worry into the exam.

Whichever you pick, check the current identification and system requirements on the booking page itself before the day. Those change, they are the sort of detail that is quietly wrong in older articles, and turning up without the right ID is a wasted fee rather than a rescheduled exam.

What happens when your certification expires

Worth understanding before you sit anything, because it shapes how much a certificate is really worth to you.

AWS certifications are valid for three years. After that the credential lapses — it does not vanish from your history, but it stops being current, and a recruiter checking will see that.

To stay current you sit the exam again, or you pass a higher-level one, which generally refreshes the lower certifications underneath it. The practical consequence: when you recertify, you sit whatever the current version is by then. Someone recertifying an ML Engineer certification in 2028 will be sitting C02 or its successor, not the version they originally passed. That is normal, and it is also the honest reason exam updates are not the disaster they first appear — everybody re-sits eventually.

Two things worth checking on AWS's own site rather than taking from any blog, including this one: the current recertification policy and whatever discount or voucher arrangement applies. Both have changed over the years and both are the sort of detail that quietly goes stale in an article. We are not quoting numbers here for exactly that reason.

Ethan's view on the three-year cycle, since it is worth having an opinion about: treat the expiry as a feature, not a tax. A cloud certification that never expired would be worthless within five years, because the platform it describes would have moved underneath it. The renewal is what keeps the credential meaning something — and it is why the courses that never update are so much worse than they look.

Questions people actually ask about this

When is the last day to take the MLA-C01 exam?

28 September 2026 is the last day to take the current version in English. Registration for the beta of the updated version opens on 1 September 2026, and beta delivery begins on 29 September 2026. If you have been studying for the current version, book before that cutoff so the exam you sit is the one you prepared for.

Should I rush to take MLA-C01 before it retires?

Only if you are nearly ready. If you have been studying for weeks and could sit it comfortably in the next few, book it, because your preparation matches that version and study material for it is mature. If you are starting from scratch now, do not rush a half-prepared attempt at a retiring exam. Aim at the new version instead, where your study time is spent on the syllabus that will still exist next year.

What is new in the MLA-C02 exam?

It brings the exam in line with how machine learning work is actually done now. It adds building and deploying generative AI solutions, fine-tuning foundation models, retrieval-augmented generation, orchestrating AI agents and complex workflows, expanded Amazon Bedrock coverage, and responsible AI across both traditional and generative workloads. No new domains were added; the existing ones were updated.

How much does the MLA-C02 beta exam cost?

The beta is 85 questions in 170 minutes for 75 US dollars, against 65 questions in 130 minutes for 150 dollars on the current version. So the beta is half the price for a longer, larger exam. The trade is that you wait longer for your result, because beta scores are held until the scoring model is finalised, and study material for a brand new version is thin.

When will the standard MLA-C02 exam be available?

The standard version is expected in early 2027, after the beta period. The beta runs in English only, and at general availability the exam is offered in English, Japanese, Korean and Simplified Chinese. If you need to sit it in a language other than English, the beta is not an option for you and waiting for general availability is the plan.

Why is my SysOps Administrator certification now called CloudOps?

It is not. AWS renamed the certification to CloudOps Engineer and updated the exam, but an existing SysOps Administrator certification stays valid to its own expiry date and does not rename itself. You only hold the CloudOps Engineer title by passing the new exam. Your existing credential is not diminished, it simply keeps the name it was issued under.

What changed in the CloudOps Engineer exam?

Containers are now in scope, roughly twenty-five further AWS services were added, and there is greater emphasis on multi-account and multi-Region architectures along with more automation and infrastructure as code. In short it moved from managing servers towards operating modern distributed systems, which is a fair description of how the job itself has changed.

What happened to the AWS Machine Learning Specialty certification?

It has been retired and its ground was split in two. The AI Practitioner covers the foundational and conceptual side for people who need to understand and discuss AI, and the Machine Learning Engineer Associate covers the practical build-and-deploy side. If a course is still selling preparation for the specialty exam, that is a strong signal nobody has maintained it.

Which AWS certification should I take in 2026?

If you are new to cloud entirely, start with Cloud Practitioner. If you want to understand and talk about AI without building it, take the AI Practitioner. If you build and deploy models, take the Machine Learning Engineer Associate. If you operate and maintain workloads, take CloudOps Engineer. They are exams for different jobs rather than a ladder you climb in order.

How do I know if a course is teaching a retired exam?

Check the exam code it names, not the course title. If it says MLS-C01 or SOA-C02, the exam no longer exists and the material has not been maintained. If it says MLA-C01 and you are reading this after September 2026, the same applies. The code takes ten seconds to check and it is the cheapest quality test available before you spend money.

Is AWS's own coming soon page reliable for exam changes?

Not entirely. Checked in August 2026 it still listed only the 2025 operations certification change and said nothing about the machine learning update happening weeks later. The AWS Training and Certification blog and the individual certification pages were both current. Check the certification page for the specific exam you are booking rather than relying on the summary page.

Does a retiring exam make my certification worthless?

No. A certification you already hold stays valid until its stated expiry date, normally three years from when you passed. Exams retire, credentials do not retire with them. What changes is that when you come to recertify, you will be sitting whatever the current version is by then, which is normal and expected.

Should I take a beta AWS exam at all?

It suits people who are confident and patient. You pay roughly half price, you sit more questions over a longer session, and you wait considerably longer for a result. You also study without the mature practice material that exists for an established version. If you need the credential by a specific date for a job or a client requirement, the beta is the wrong choice.

Do certification changes mean my study notes are wasted?

Mostly not. Exam updates typically keep the existing domains and refresh what sits inside them, so the fundamentals you learned still apply. In this case the machine learning update adds generative AI and agent topics on top rather than replacing the traditional material. Treat your notes as a base to extend, not something to throw away and restart.

Will an AWS certification get me a job?

On its own, no, and anyone promising otherwise is selling something. What it does is give you a syllabus written by people who know the material, a deadline that makes you finish, and a credential a recruiter recognises. The people who convert certifications into work are the ones who also built something they can talk about. The certificate opens a conversation; the project wins it.

What the lad did in the end

Booked the AI Practitioner for a hundred dollars, because most of what he had learned in four months mapped onto it, and it gave him something finished rather than something abandoned.

Then started on the ML Engineer material aimed at the new version, on the reasoning that if he was going to spend another few months studying, it might as well be for the exam that exists next year.

"He was fine about it in the end," Jake said. "Annoyed at the course, not at AWS."

"That's the right way round. The exam changing is the system working. The course still selling is the part nobody polices."

So: check the code before you pay. And if MLA-C01 is the one you have been preparing for, you have until 28 September — which is not long, but it is enough if you are nearly there.

A note on dates. Published 9 August 2026. Every figure here — the September dates, the question counts, the durations and the prices — was read from AWS's own pages on the day of writing. Certification details move, and a wrong date here could cost you a booking fee, so check the official page before you pay. This page will be updated after 28 September, when the MLA-C01 deadline passes and the beta is under way; the answer to "which exam should I take" changes on that date, so this page will be updated then. If you spot something that has moved before we do, please tell us through the contact page — a reader who catches it first saves everyone who arrives after them, and on a subject where people are about to spend a few hundred dollars, that genuinely matters.