Study guide · Exam overview
The format, passing score, and four content domains of the AWS Certified Machine Learning Engineer – Associate exam, including the generative AI work that replaced the retired Machine Learning Specialty exam.
Exam version notice: AWS Certified Machine Learning – Specialty (MLS-C01) retired on March 31, 2026. This overview follows the AWS Certified Machine Learning Engineer – Associate exam guide for MLA-C02, reviewed in October 2026. English delivery is the beta that started September 29, 2026. General availability in all languages begins January 14, 2027. Items tagged Extra are useful context the current guide removed or placed outside the target candidate’s job. If service names are still new, start with the AI Practitioner overview and the Cloud Practitioner overview.
MLA-C02 checks whether you can prepare data, choose and train or customize a model, deploy it, and keep the running system observable, affordable, and locked down. The exam guide says the target candidate has at least one year with Amazon SageMaker AI, Amazon Bedrock, and related AWS services, plus at least one year in a role such as backend development, DevOps, data engineering, or data science. The candidate is expected to know both traditional machine learning and generative AI.
A question usually gives a constraint: a feature that will not exist at inference, a latency cap, a document corpus, an agent that must call an API, a drift alarm, or a bucket the endpoint role should not read. Two options often both “work.” The better answer meets the constraint with the managed path the stem already allows. Read the noun the stem cares about: the data shape, the model you are allowed to change, the traffic pattern, or the control that failed.
The exam guide also names work the target candidate is not expected to do: design a full end-to-end AI strategy, set organization-wide ML standards, integrate a wide array of unfamiliar tools, or work deeply in two or more ML domains at once. A question that sounds like a strategy memo is usually asking for one implementation step.
Figures below come from the MLA-C02 exam guide and the AWS certification page reviewed in October 2026.
| Specification | MLA-C02 |
|---|---|
| Level | Associate. There is no required prerequisite certification. |
| Current English delivery | Beta, from September 29, 2026. The certification page lists the beta registration code as ME1-C02. The exam guide version name is MLA-C02. |
| Beta length and price | 85 questions, 170 minutes, 75 USD. English only until general availability. |
| Scored form in the exam guide | 50 questions affect the score. 15 are unscored and are not identified. Question types are multiple choice and multiple response. |
| Passing score | 720 on a scaled range of 100–1,000. Scoring is compensatory. AWS says this pass-or-fail designation does not apply to the beta. |
| General availability | January 14, 2027, in all languages. MLA-C01 retires in every language that day. The last day for MLA-C01 in English was September 28, 2026. |
| Recommended experience | At least 1 year with SageMaker AI, Bedrock, and other ML engineering services, and at least 1 year in a related role, including traditional ML and generative AI. |
| Validity | Three years. Confirm the current recertification options with AWS. |
Unanswered questions are scored as incorrect. There is no penalty for guessing, so answer every question. A scaled score of 720 is not the same thing as 72 percent of the questions. Confirm the general-availability duration and fee on the AWS certification site, because the published 85-question, 170-minute, 75 USD figures are the beta figures.
These percentages are the share of scored content in the official exam guide. The 200 practice questions on this site follow the same weights.
The largest domain. Get the bytes in, shape them, and prove they are fit to learn from.
Pick the approach, train or customize it, and measure it.
Host the model or agent and ship it through a pipeline.
Notice the failure, pay for the right capacity, and lock the artifact down.
The exam guide publishes an in-scope list. Learn those services first. The same map is in the core notes and flashcards.
| Need | Prefer | Remember |
|---|---|---|
| Shared features for train and serve | SageMaker Feature Store | A one-off CSV in S3 can drift from the live request. |
| Similar documents or images | Embeddings plus OpenSearch Service, RDS with pgvector, or S3 | A Glue catalog is a schema, not a vector index. |
| A foundation model you do not train | Amazon Bedrock | Compare quality, latency, and cost before you fine-tune. |
| A classical model with a built-in algorithm | Amazon SageMaker AI | Script mode is for a training loop the built-in option does not cover. |
| Forms, speech, text, or images without training | Textract, Transcribe, Comprehend, or Rekognition | Polly speaks. It does not read a scan. |
| Retrieval over your files | An Amazon Bedrock knowledge base | Refresh the index when the source changes, then test retrieval. |
| An agent platform | Amazon Bedrock AgentCore | Store agent state and watch tool failures. |
| A repeatable model release | SageMaker Model Registry and AWS CodePipeline | Roll back to the last approved version. Do not edit production in place. |
| Unsafe model output | Amazon Bedrock Guardrails | A larger model does not replace a content filter. |
| Who changed the endpoint | AWS CloudTrail | CloudWatch shows the symptom. Config shows the resource shape. |
| Extra Compile a model for a phone | Leave SageMaker Neo out of the answer | The MLA-C02 comparison removed edge optimization with SageMaker Neo. |
| Extra Shrink a model by pruning | Prefer early stopping, data, or a smaller model choice | Pruning and compression were removed from the training task. |
| Extra Design the company’s AI strategy | Answer the implementation step in the stem | End-to-end strategy and org-wide standards are out of scope for the target candidate. |
Multiple choice has one correct response and three distractors. Multiple response has two or more correct responses out of five or more options. The stem tells you how many to choose. Select every response that completes the statement, and nothing extra. The MLA-C02 exam guide does not list ordering or matching. Those formats were on MLA-C01.
Distractors are real services that solve a nearby problem. If the need is “replay this hour,” a store that only delivers onward is the nearby wrong answer. If the need is “the manual must be cited,” a larger foundation model with no index is the nearby wrong answer. If the need is “the endpoint role should not read another team’s bucket,” AdministratorAccess is broader than the stem asked for.
Words such as MOST, BEST, and LEAST are the constraint. MOST cost-effective still has to meet the latency number. LEAST operational overhead usually means the managed service that already fits. There is no published penalty for a wrong guess. A blank answer is incorrect.
A practical sequence once you can already name the core AWS services.
A table, a stream, a document corpus, or a prompt-and-response file. The prep tool follows that sentence.
A managed API or a Bedrock model may already hit the metric. Fine-tune when the gap is real and the pairs exist.
A wrong citation is a chunk, an index, or a refresh. A fluent contradiction of a good chunk is the generator.
Register the model, version the prompt, test, then shift traffic. Keep the previous version for rollback.
Use the four decks until you can answer before the card flips. Revisit misses the next day.
Work through original questions at about two minutes each, review every explanation, and return to the notes on domains below 75 percent. Aim for a steady 85 percent before you book.
The overview, core notes, flashcards, and practice exams are all published.
Exam format, the 720 passing score, domain weights, and a study plan.
Reference notes for data, models, deployment, and operations.
40 recall cards in four decks, plus memory notes for models, retrieval, and controls.
200 original questions with custom exams, explanations, and a score report by domain.
AWS retired AWS Certified Machine Learning – Specialty (MLS-C01). The last day to take that exam was March 31, 2026. A credential already earned stays active for three years from the date it was earned. The current exam for the machine learning engineer role is AWS Certified Machine Learning Engineer – Associate (MLA-C02).
The exam guide names the updated exam MLA-C02. As of October 2026, English candidates sit the MLA-C02 beta. The AWS certification page lists that beta registration code as ME1-C02. General availability in all languages begins January 14, 2027, and MLA-C01 retires in every language on that date. Confirm the code on the AWS certification page before you book.
The exam guide reports a scaled score from 100 to 1,000. The minimum passing score is 720, and you pass the exam as a whole rather than each domain. AWS states that this pass-or-fail designation does not apply to the beta version. A scaled score is not the percentage of questions you answered correctly.
The English beta, in delivery from September 29, 2026, has 85 questions and a 170-minute limit, and AWS lists the beta price at 75 USD. The exam guide describes 50 scored questions plus 15 unscored questions, and question types of multiple choice and multiple response. Confirm the general-availability length, price, and languages with AWS. Unanswered questions are incorrect, and there is no penalty for guessing.
There is no required prerequisite certification. AWS describes the target candidate as someone with at least one year using Amazon SageMaker AI, Amazon Bedrock, and other AWS ML engineering services, plus at least one year in a related role, and experience with both traditional machine learning and generative AI.
No. The practice questions on this site are original. They are not AWS exam items, and a practice score does not predict a result.
Service names and prices change. Use the current AWS exam guide and the AWS certification site as the authority when they differ from these notes.
Useful companions while you study.
Convert structured data between JSON and YAML with formatting and validation.
Convert CSV spreadsheet data to JSON or turn JSON arrays into downloadable CSV files.
Encode text and files to Base64 or decode Base64 data with UTF-8 support.
Compare two text blocks and highlight added, removed, and changed content.
Exam versions, objectives, prices, languages, and retirement dates can change. Verify the details on the AWS certification page before you register.
Certicab provides general informational, educational, and convenience resources. Calculations, conversions, estimates, and learning materials may contain errors or become outdated. Financial, tax, medical, legal, and travel information is not professional advice. Verify important results and current requirements with qualified professionals or authoritative sources. Protect sensitive files and personal information, review each tool’s privacy limitations, and use only content you are authorized to process. You are responsible for how you use and share results. Study resources are independent and do not guarantee exam success or imply certification-provider endorsement. Amazon Web Services, AWS, and related marks are trademarks of Amazon.com, Inc. or its affiliates. Certicab is not affiliated with, endorsed by, or sponsored by Amazon.
Certi