AWS Machine Learning Engineer Associate MLA-C01 to MLA-C02: What’s Changing and What to Do Now
AWS Certified Machine Learning Engineer – Associate has moved from MLA-C01 to MLA-C02. According to the AWS Training and Certification September 2026 announcement, the last day to sit MLA-C01 in English was 28 September 2026, and the English-only MLA-C02 beta opened for delivery on 29 September 2026. General availability in all languages begins on 14 January 2027. On the same date, MLA-C01 is retired in every language.
The four-domain structure stays the same. What changes is the scope: MLA-C02 adds generative AI, retrieval-augmented generation (RAG), agentic AI, and foundation models (FMs) to the traditional ML engineering skills. This article covers what AWS has officially published, how the domain weights shift, who is affected, and how to adjust your study plan.
Key dates
- 1 September 2026: MLA-C02 beta registration opens (English only), and the new exam guide is released
- 28 September 2026: last day to take MLA-C01 in English
- 29 September 2026: MLA-C02 beta delivery begins
- 14 January 2027: MLA-C02 GA delivery begins in all languages, and MLA-C01 is retired in all languages
The original MLA-C02 announcement from July 2026 said only “early 2027” for the standard version. The 14 January 2027 date comes from the later September post. At the time of writing, the official certification page still lists GA registration and delivery as “TBD”, so check it again before you book.
Exam format: beta vs C01
| Detail | MLA-C02 beta | MLA-C01 |
|---|---|---|
| Duration | 170 minutes | 130 minutes |
| Questions | 85 (multiple choice and multiple response) | 65 (multiple choice and multiple response) |
| Price | $75 USD (beta pricing) | $150 USD |
| Languages | English only | Japanese, Korean, Simplified Chinese (English ended 28 Sep 2026) |
| Delivery | Pearson VUE test center or online proctored | Pearson VUE test center or online proctored |
| Validity | 3 years | 3 years |
The beta has more questions and more time because it includes extra items that AWS uses for statistical evaluation. AWS says beta results are typically available within five business days. The certification page lists the beta under exam code ME1-C02, which is worth knowing when you search for it on Pearson VUE.
For the standard version, the MLA-C02 exam guide lists 50 scored and 15 unscored questions, a scaled score from 100 to 1,000, and a minimum passing score of 720, using compensatory scoring. AWS has not yet published the GA duration or price on the certification page, so we won’t guess them here.
Domain comparison: C01 vs C02
Both weightings come from AWS’s official MLA-C01 vs MLA-C02 comparison.
| MLA-C01 domain | C01 | MLA-C02 domain | C02 | Change |
|---|---|---|---|---|
| 1. Data Preparation for Machine Learning (ML) | 28% | 1. Data Preparation for ML and AI | 28% | 0% |
| 2. ML Model Development | 26% | 2. ML Model and Foundation Model (FM) Development | 24% | −2% |
| 3. Deployment and Orchestration of ML Workflows | 22% | 3. Deployment and Orchestration of ML and AI Workflows | 24% | +2% |
| 4. ML Solution Monitoring, Maintenance, and Security | 24% | 4. Operating, Monitoring, and Securing ML and AI Solutions | 24% | 0% |
The weights barely move. Model development drops two points and deployment and orchestration gains two. The real change is in the domain titles, which now say “ML and AI“, and in the task statements underneath them. Each of the twelve C01 task statements maps directly onto a C02 task with the same number.
What was added
The comparison page lists new skills in every task. Grouped by theme:
- GenAI data preparation: configuring vector databases (for example, OpenSearch Service, Amazon RDS with pgvector, Amazon S3); embedding models; chunking and metadata extraction for RAG; masking and redaction; and preparing data for FM fine-tuning, continuous pre-training, and distillation.
- Foundation models and RAG: choosing FMs in Amazon Bedrock, fine-tuning strategies, RAG architecture patterns, tuning retrieval and embeddings, and the trade-offs between performance, latency, and cost.
- AI evaluation: human-in-the-loop evaluation; NLP metrics such as BLEU, ROUGE, and BERTScore; LLM-as-a-judge; and monitoring RAG retrieval accuracy.
- Agentic AI: deploying agents and agent communication protocols, managing agent state, building agentic workflow infrastructure, automating agent deployment pipelines, and versioning agents.
- Operations and cost: Amazon Bedrock Prompt Management, Bedrock Custom Model Import, knowledge base refresh pipelines, GPU scaling, monitoring token usage and FM inference costs, and monitoring agent failures.
- Security and responsible AI: scanning CI/CD pipelines for vulnerabilities (Amazon CodeGuru, Amazon Inspector), Amazon Bedrock API keys vs IAM credentials, and Amazon Bedrock Guardrails.
What was removed
AWS lists these C01 items as deleted from C02:
- Configuring data to load into training resources (Amazon EFS, Amazon FSx)
- Fine-tuning with custom datasets as a standalone item (Amazon Bedrock, SageMaker JumpStart) and reducing model size through pruning or compression
- Optimizing models for edge devices (SageMaker Neo)
- Bring your own container (BYOC) with SageMaker
- Monitoring infrastructure with Amazon EventBridge events, and troubleshooting capacity issues such as provisioned concurrency, service quotas, and auto scaling
Fine-tuning hasn’t disappeared. It comes back as FM fine-tuning strategy, data preparation, and deployment automation for fine-tuned model versions.
Who is affected
- English candidates partway through C01 prep: the C01 English window has closed. You can take the beta now (at $75, with 85 questions in 170 minutes) or wait for the standard version from 14 January 2027. Either way, add the GenAI, RAG, and agent topics above to your plan.
- Beta takers: expect a longer sitting with unscored items mixed in, and a wait of a few business days for results. Prepare against the C02 exam guide, not C01 materials.
- Japanese, Korean, and Simplified Chinese candidates: MLA-C01 stays available in these languages until MLA-C02 GA on 14 January 2027. If you’re nearly ready, that’s still an option.
- Current holders: your certification stays active through its original expiration date. You recertify by passing the latest version of the exam.
A practical study plan for MLA-C02
- Weeks 1–2: baseline. Read the C02 exam guide and the comparison page. Test yourself on classic SageMaker AI topics such as data preparation, training, tuning, endpoints, and MLOps pipelines. These still make up most of the exam.
- Weeks 3–4: GenAI and RAG. Build a small Amazon Bedrock knowledge base. Try chunking strategies, embeddings, a vector store, and reranking, then evaluate the results with RAG metrics and LLM-as-a-judge.
- Week 5: agents and operations. Deploy a simple agent, version it, and monitor its tool failures and token costs. Add Guardrails and Prompt Management.
- Week 6: timed practice. Sit full-length mock exams under time pressure (170 minutes if you’re taking the beta), review every wrong answer against the matching task statement, and give the most time to domains 1–3, which carry 76% together.
For scenario practice, try the AWS Machine Learning Engineer Associate practice exam on Testimea, or start with the ML Engineer Associate practice exam trial. If you’re new to AWS AI, the AWS AI Practitioner practice exam makes a gentler first step.
FAQ
Can I still take MLA-C01 in English?
No. The last English delivery date was 28 September 2026. C01 continues in Japanese, Korean, and Simplified Chinese until 14 January 2027.
Are there new domains in MLA-C02?
No. AWS confirms the domain structure is unchanged. Only the titles, weights (±2%), and task-level skills changed.
How much does the beta cost and how long is it?
$75 USD, 85 questions, 170 minutes, in English only.
What is the passing score?
The C02 exam guide sets a minimum of 720 on a 100–1,000 scale.
Does my MLA-C01 certification become invalid?
No. It stays active until its original expiration date, three years from when you earned it.
Bottom line
MLA-C02 keeps the four domains and nearly the same weights, but it now expects ML engineers to handle foundation models, RAG, agents, and Amazon Bedrock alongside SageMaker AI. English candidates can take the beta now or wait for GA on 14 January 2027. Study from the C02 guide either way. Ready to practise? Try the ML Engineer Associate trial or the full AWS Machine Learning Engineer Associate practice exam.
