Blog Post
AWS Machine Learning Engineer exam updated from MLA-C01 to MLA-C02. New GenAI, foundation model, and agentic AI content. Exam dates, domain changes, and study tips.
Checking sign-in status...
The AWS Certified Machine Learning Engineer - Associate exam has been updated from MLA-C01 to MLA-C02. The new version adds generative AI, foundation models, large language models, and agentic AI to the core scope. If you were preparing for MLA-C01, the material you studied covers only part of what MLA-C02 now tests.
The English MLA-C01 exam retired on September 28, 2026. MLA-C02 beta registration opened September 1, 2026. Beta delivery began September 29, 2026. General availability is expected January 14, 2027.
This guide covers what changed, the new domain weights, new skills, and how to prepare.
| Date | Event |
|---|---|
| September 1, 2026 | MLA-C02 beta registration opens (English only) |
| September 1, 2026 | MLA-C02 exam guide released |
| September 28, 2026 | Last day to take MLA-C01 in English |
| September 29, 2026 | MLA-C02 beta delivery begins |
| January 14, 2027 | MLA-C02 general availability (all languages) |
The MLA-C01 exam remains available in Japanese, Korean, and Simplified Chinese during the beta period. Once MLA-C02 reaches general availability, MLA-C01 retires in all languages.
| Detail | MLA-C02 Beta | MLA-C01 Standard |
|---|---|---|
| Duration | 170 minutes | 130 minutes |
| Questions | 85 | 65 (50 scored + 15 unscored) |
| Price | $75 USD | $150 USD |
| Passing score | No pass/fail during beta | 720/1000 |
| Languages | English only | English, Japanese, Korean, Simplified Chinese |
| Question types | Multiple choice and multiple response | Multiple choice, multiple response, ordering, matching |
The beta exam is longer and cheaper. Beta results are not scored immediately. AWS collects question quality data before finalising scores. If you pass the beta, the certification is valid for three years.
Ordering and matching question types were removed from MLA-C02. The exam now uses only multiple choice and multiple response.
The four-domain structure is preserved. No new domains were added. But two domains were renamed, and the weights shifted.
| Domain | MLA-C01 Weight | MLA-C02 Weight | Change |
|---|---|---|---|
| D1: Data Preparation for ML | 28% | 28% | Renamed to Data Preparation for ML and AI |
| D2: ML Model Development | 26% | 24% | Renamed to ML Model and Foundation Model Development |
| D3: Deployment and Orchestration of ML Workflows | 22% | 24% | Renamed to Deployment and Orchestration of ML and AI Workflows |
| D4: ML Solution Monitoring, Maintenance, and Security | 24% | 24% | Renamed to Operating, Monitoring, and Securing ML and AI Solutions |
Two points moved from Domain 2 to Domain 3. The twelve-task skeleton is otherwise preserved.
The updated exam reflects how the ML engineer role has broadened. The key additions are grouped into four categories.
Building and deploying GenAI solutions. Fine-tuning foundation models. Implementing retrieval-augmented generation (RAG) architectures. Configuring vector databases with OpenSearch Service or RDS with pgvector. Preparing documents for RAG with chunking strategies and metadata extraction.
Orchestrating AI agents and complex workflows. Deploying and configuring agents with service and tool integration. Managing agent state and communication protocols. Building agentic workflow infrastructure. Automated agent deployment and versioning.
Selecting, customizing, and operationalizing large language models. Evaluating foundation model deployment options. Bringing models built outside AWS via SageMaker AI or Bedrock Custom Model Import. NLP evaluation metrics including BLEU, ROUGE, and BERTScore. LLM-as-a-judge evaluation frameworks.
Expanded coverage of Bedrock capabilities. Bedrock Knowledge Bases, Prompt Management, Guardrails, Custom Model Import, and Model Evaluation. Bedrock AgentCore for agent orchestration. Credential types for foundation model access including Bedrock API keys and IAM.
| Domain | New Skills |
|---|---|
| D1: Data Preparation | Vector database configuration, multimodal data ingestion (text, image, audio), embedding models, RAG document preparation, data masking and anonymization |
| D2: Model and FM Development | FM selection from Bedrock, fine-tuning strategies, RAG architecture patterns, performance-latency-cost tradeoffs, NLP metrics (BLEU, ROUGE, BERTScore), LLM-as-a-judge |
| D3: Deployment and Orchestration | Evaluating FM deployment options, Bedrock Custom Model Import, agent deployment and configuration, RAG configuration with reranking, Bedrock Knowledge Bases, GPU scaling, automated agent versioning |
| D4: Operations and Security | Agent performance monitoring, FM cost evaluation, AI-specific cost patterns (token usage, embedding compute), Bedrock Guardrails, code and image vulnerability scanning |
Nearly every addition sits in the Bedrock-centered generative AI, RAG, and agent space.
| Removed Skill | Why It Matters |
|---|---|
| Configuring data loading to training resources (EFS, FSx) | Storage provisioning for ML training is less tested |
| Fine-tuning pre-trained models on custom datasets (JumpStart) | Reframed in the FM context via Bedrock |
| Reducing model size (data type changes, pruning, compression) | Traditional model compression is out of scope |
| Optimizing for edge devices (SageMaker Neo) | Edge ML is out of scope |
| Bring your own container on SageMaker | Container customization is out of scope |
| Infrastructure monitoring with EventBridge | Folded into broader monitoring skills |
| Troubleshooting capacity issues (provisioned concurrency, service quotas) | De-emphasized |
Fine-tuning did not disappear. It was redefined in the foundation model context. What got lighter is the traditional ML periphery: SageMaker Neo, BYOC, and model compression.
| Change | Services |
|---|---|
| Added in MLA-C02 | Amazon Bedrock AgentCore, AWS CodeConnections, Amazon Inspector |
| Renamed | Amazon SageMaker → Amazon SageMaker AI |
| Removed in MLA-C02 | Amazon A2I, Amazon Fraud Detector, Amazon Kendra, Lookout for Equipment/Metrics/Vision, Amazon Mechanical Turk, Amazon Q, AWS Chatbot, AWS Serverless Application Repository, CloudWatch Logs (folded into CloudWatch) |
The removal of Amazon Q and Kendra reflects RAG's center of gravity moving to Bedrock Knowledge Bases. If you were studying Kendra for RAG, switch to Bedrock Knowledge Bases.
The MLA-C02 target candidate now needs:
The MLA-C01 profile required SageMaker experience. MLA-C02 adds Bedrock experience as a baseline expectation.
If you were studying for MLA-C01: You have a head start on Domains 1 and 4. The traditional ML content is still tested. Focus your additional study on Bedrock, RAG architectures, agentic AI, foundation model selection, and the new evaluation metrics (BLEU, ROUGE, BERTScore).
If you are starting fresh: Begin with SageMaker fundamentals and traditional ML. Then layer on Bedrock and generative AI. The exam tests both. Do not skip the traditional ML content because GenAI is newer.
Hands-on priorities:
ExamOS practice: Use the Daily Sprint for domain-level drilling. Your Readiness Score updates after every session and recommends booking when you cross the target threshold. Focus additional Mock Exam practice on Domains 3 and 4, which carry 48% of the exam combined.
The MLA-C01 exam retired in English on September 28, 2026. It remains available in Japanese, Korean, and Simplified Chinese during the MLA-C02 beta period. Once MLA-C02 reaches general availability on January 14, 2027, MLA-C01 retires in all languages.
The passing score is 720/1000 for the standard exam. During the beta period, there is no pass or fail result. AWS collects question quality data before finalising scores. If you pass the beta, the certification is valid for three years.
Yes. The target candidate profile now recommends at least 1 year of experience with Amazon Bedrock in addition to SageMaker AI. The exam tests foundation model selection, RAG architectures, Bedrock Knowledge Bases, Guardrails, and agent deployment.
Data Preparation for ML and AI is the heaviest domain at 28%. Domains 3 and 4 together carry 48% of the exam weight. Domain 3 increased from 22% to 24%, reflecting the growing importance of deployment and orchestration skills for AI workloads.
Exam fees and dates are sourced from AWS official announcements and are subject to change. Verify current details on the AWS Certification website before booking.
Build MLA-C02 readiness with the ExamOS Daily Sprint. Seven questions in seven minutes, adapting to your weak areas every day. Start with the AWS ML Engineer track and identify your gaps before exam day.