AWS Certified Generative AI Developer - Professional
315 practice questions
Last reviewed: April 2026
Personal notes and resource links for your study journey
Filter by Certification
The AWS Certified Generative AI Developer Professional (AIP-C01) is a professional-level credential focused on building, integrating, and operating production-grade generative AI applications on AWS — primarily on Amazon Bedrock, with SageMaker, Lambda, OpenSearch, and Knowledge Bases as supporting services. It targets experienced developers and ML engineers who design retrieval-augmented generation (RAG) systems, agentic workflows, and tool-using foundation-model applications. Expect deep scenario questions on prompt engineering, embeddings, vector search, guardrails, evaluation, cost optimization, and responsible-AI controls. Unlike the foundational AIF-C01, AIP-C01 assumes the candidate writes code and has shipped Bedrock-based features. The exam is multiple-choice and multiple-response, no hands-on labs.
Largest domain at 31%. Bedrock model selection, embedding pipelines, vector stores (OpenSearch Serverless, Aurora pgvector, Kendra), Knowledge Bases, fine-tuning vs. continued pre-training, and data governance for LLM training data.
Building RAG, agents, and tool-using applications. Bedrock Agents, Action Groups, Lambda integrations, and orchestrating multi-step LLM calls. Common stumbling block: knowing when an Agent vs. a custom orchestration is appropriate.
Bedrock Guardrails, prompt-injection mitigations, PII redaction, IAM and KMS for model access, and content moderation patterns. High-density questions despite the modest 20% weight.
Inference cost control (provisioned vs. on-demand throughput), caching, model distillation, latency tuning, and choosing smaller models where appropriate. Often missed: when to use cross-region inference profiles.
Evaluation frameworks (Bedrock model evaluation, human evaluation, LLM-as-judge), drift, hallucination detection, and debugging RAG retrieval failures. Smallest domain (11%) but punishes shallow study.
Services you'll encounter on the exam and why each one matters.
Fully managed service exposing foundation models from Anthropic, Meta, Mistral, AI21, Cohere, Stability AI, and Amazon Titan through a single API with streaming, batch, and provisioned throughput.
Why it's on the exam: Bedrock is the substrate for every Domain 1 (Foundation Model Integration) scenario — model selection, inference parameters, provisioned vs. on-demand throughput, and streaming response handling.
Managed RAG built on S3 documents and a pluggable vector store (OpenSearch Serverless, Aurora pgvector, Pinecone, MongoDB Atlas, Redis Enterprise) with chunking, embeddings, and citation support.
Why it's on the exam: Domain 2 (Implementation and Integration) tests when to ground a model in private data via RAG instead of fine-tuning, including chunking strategy, vector-store choice, and citation propagation.
Policy layer filtering harmful content, blocking denied topics, redacting PII, applying contextual-grounding checks, and surfacing word/topic filters across input and output.
Why it's on the exam: Domain 3 (AI Safety, Security, and Governance) is largely a Guardrails questionnaire — content filter strengths, denied-topic config, contextual grounding, and PII anonymization patterns.
Orchestration layer that lets a foundation model call APIs, query knowledge bases, and chain multi-step actions via tool use, with action groups backed by Lambda and OpenAPI/function schemas.
Why it's on the exam: Domain 2 scenarios on multi-step reasoning, tool-calling agents, and Lambda-backed action groups name Agents as the AWS-native orchestrator; expect distinctions vs. raw inference or Step Functions.
Fine-tuning and continued pre-training of foundation models on private S3 datasets, producing custom models deployable via Provisioned Throughput.
Why it's on the exam: Domain 1 questions test the fine-tune vs. RAG vs. prompt-engineering decision tree; expect specific scenarios where continued pre-training or instruction fine-tuning is the right answer.
Managed automatic and human-in-the-loop model evaluation — built-in metrics (accuracy, robustness, toxicity) plus custom rubrics scored by LLM-as-judge or human workers.
Why it's on the exam: Domain 5 (Testing, Validation, and Troubleshooting) tests how to quantify quality regressions between models or prompt variants — Bedrock Evaluations is the named mechanism.
Catalogue of pre-trained foundation and task-specific models with one-click deployment to SageMaker endpoints, transfer-learning notebooks, and proprietary-model marketplace.
Why it's on the exam: Domain 2 distinguishes "host on SageMaker endpoint" (JumpStart) from "call Bedrock API" — questions test latency, cost, and customization tradeoffs between the two.
Packaged generative-AI assistant — Q Developer for IDE/Console coding workflows, Q Business for enterprise RAG over connected data sources, Q in QuickSight for BI Q&A.
Why it's on the exam: Domain 2 tests when to consume Q (packaged) versus build on Bedrock (bespoke); Q Business specifically appears in enterprise-RAG scenario questions.
Serverless OpenSearch with a dedicated vector-search collection type — the default Bedrock Knowledge Base backend and the most-tested AWS-native vector store.
Why it's on the exam: Domain 1 RAG scenarios assume OpenSearch Serverless as the default vector store — expect questions on OCU sizing, capacity-vs-cost, and approximate-nearest-neighbor index choices.
Aurora PostgreSQL with the pgvector extension, supported by Bedrock Knowledge Bases as an alternative vector store with transactional semantics and SQL access.
Why it's on the exam: Domain 2 vector-store-selection questions contrast Aurora pgvector (existing SQL stack, transactional) against OpenSearch Serverless (pure-vector, serverless OCU).
Object storage holding Knowledge Base source documents, fine-tuning training corpora, evaluation datasets, and model invocation logs.
Why it's on the exam: Every Domain 1 data-management scenario passes through S3 — chunking source layout, lifecycle on raw documents, and IAM access on KB buckets are recurring exam topics.
Serverless compute for Bedrock Agent action groups, RAG preprocessing, post-inference response shaping, and S3-event-driven ingestion into Knowledge Bases.
Why it's on the exam: Domain 2 action-group questions test the Lambda contract (request/response schema) that Bedrock Agents invoke; Domain 4 also tests Lambda as the cost-efficient inference glue.
Serverless workflow orchestrator with native Bedrock InvokeModel integration, ideal for multi-call prompt chains, parallel evaluations, and human-approval loops.
Why it's on the exam: Domain 2 distinguishes Step Functions (deterministic multi-step orchestration) from Bedrock Agents (model-driven tool calling); knowing when each is correct is a recurring distractor.
Managed HTTP/REST/WebSocket front door for Bedrock-backed inference endpoints, with throttling, API keys, JWT authorization, and usage-plan tiering.
Why it's on the exam: Domain 2 productionization scenarios test exposing RAG and agent endpoints externally — API Gateway is the named entry point, paired with Lambda or direct Bedrock integration.
ML platform used in AIP-C01 mainly for hosting JumpStart foundation models on real-time/serverless/async endpoints when Bedrock's managed API isn't the right fit.
Why it's on the exam: Domain 2 endpoint-deployment questions test real-time vs. serverless vs. async-inference choices on SageMaker; cost and cold-start tradeoffs vs. Bedrock are the typical pivot.
Managed NLP with a PII-detection API that returns entity spans and types, plus targeted PII redaction — usable as a pre-Guardrails filter or post-inference scrubber.
Why it's on the exam: Domain 3 PII-handling questions test layered defenses: Comprehend for detection, Guardrails for blocking — knowing where each belongs in the pipeline is exam fodder.
Account-wide access control with service roles, identity-based and resource-based policies, and condition keys governing Bedrock model access, Agent execution, and KB ingestion.
Why it's on the exam: Domain 3 (AI Safety, Security, and Governance) tests least-privilege roles for Bedrock invocation, Agent action groups, and Knowledge Base ingestion — IAM is the named mechanism throughout.
Customer-managed encryption keys for Bedrock model customization data, Knowledge Base vector indexes, S3 source documents, and CloudWatch invocation logs.
Why it's on the exam: Domain 3 compliance scenarios test customer-managed-key control over fine-tuning corpora and prompt/response logs; KMS grants and key policies are the named answer.
Account-wide audit log of every API call — who invoked which Bedrock model, who updated Guardrails, who ingested data into a Knowledge Base, with optional data-event capture for inference.
Why it's on the exam: Domain 3 audit-trail scenarios cite CloudTrail as the immutable record needed for compliance reviews and incident response across Bedrock and Agent activity.
Metrics, logs, and alarms — Bedrock model-invocation logs (prompts and completions), Agent traces, Knowledge Base ingestion progress, and Lambda action-group execution metrics.
Why it's on the exam: Domain 4 (Operational Efficiency) and Domain 5 (Troubleshooting) test CloudWatch for token-cost observability, latency alarms, and tracing failed Agent action-group invocations.
$140k–$195k–$280k USD annual
AIP-C01 is a newer credential and lacks dedicated salary surveys. Range derived from adjacent GenAI / ML engineering compensation in the US market and should be treated as approximate. Senior GenAI engineers at top-tier AI labs and FAANG often exceed $400k TC. Entry-level "GenAI" roles can dip below the low end. Your mileage will vary substantially by company tier, location, and demonstrated shipped work.
Source: levels.fyi 2025–2026 GenAI / ML engineer roles (adjacent), U.S. BLS OEWS May 2024 (15-1252 software developers, 15-2051 data scientists). Figures are approximate; actual compensation depends on role, region, and experience.
GenAI engineering roles became one of the fastest-growing job families through 2024–2026 as enterprise Bedrock and LLM adoption moved from prototypes to production systems. AIP-C01 is positioned as a credible professional-level signal that a candidate can ship Bedrock-based applications, including RAG, agents, and guarded production endpoints. Recruiters at AWS-centric enterprise shops use it alongside demonstrated Bedrock projects on GitHub or production work. It pairs strongly with AIF-C01 (foundation), MLA-C01 (engineering breadth), and Solutions Architect Professional (SAP-C02) for multi-domain credibility. The cert does NOT by itself qualify candidates for ML research, foundation-model training roles, or applied-science positions — those expect deep ML fundamentals and often a graduate degree.
There are no formal prerequisites. AWS recommends at least one year of experience building applications with foundation models on AWS, plus a year of broader software engineering experience. Practical expectations include comfort with Python, REST/SDK integration, vector databases, and at least one production Bedrock or SageMaker JumpStart deployment.
The recommended path is AIF-C01 first to absorb GenAI vocabulary, then either MLA-C01 (engineering depth) or DVA-C02 (developer fluency) before tackling AIP-C01. Candidates without prior AWS exposure should expect a steep curve — many AIP-C01 questions assume baseline familiarity with IAM, VPC, Lambda, and API Gateway. A working personal RAG project with Bedrock plus OpenSearch Serverless or Knowledge Bases is the single most useful preparation artifact.
AIP-C01 is rated Professional and is one of the harder AWS associate-or-higher exams because it spans both LLM application engineering and AWS-platform depth. Plan 100–160 hours over 10–14 weeks if you already build GenAI applications professionally; 200+ hours over 16+ weeks if you are coming from a non-GenAI background. The exam is 75 scored questions in 180 minutes — multiple-choice and multiple-response, no labs.
Common stumbling blocks include the breadth of Bedrock features (Agents, Action Groups, Guardrails, Knowledge Bases, model evaluation, custom model import, cross-region inference), nuanced cost-optimization scenarios involving provisioned throughput, and subtle questions about RAG retrieval quality vs. fine-tuning tradeoffs. The exam also rewards practical familiarity with prompt-injection and data-leakage mitigations.
Initial general availability of the professional-level Generative AI Developer credential. Replaces and extends the foundational AIF-C01 with deep Bedrock and LLM-application content. Current version as of April 2026.
AIP-C01 (AWS Certified Generative AI Developer - Professional) is a a challenging, scenario-heavy exam that requires deep hands-on experience and the ability to make architectural trade-off decisions Professional-level exam. Most candidates need 150–300 hours of study spread over 3–6 months for professional and expert-level exams. These exams typically expect prior associate-level proficiency. Most candidates who score consistently above the passing threshold on practice exams pass on their first attempt.
Most candidates need 150–300 hours of study spread over 3–6 months for professional and expert-level exams. These exams typically expect prior associate-level proficiency. Time-to-pass varies widely by prior experience. Engineers with hands-on production experience in the underlying technology typically need less; candidates new to the platform should plan toward the upper end of that range.
AIP-C01 is a recognized credential in the AWS ecosystem and signals validated knowledge to employers, recruiters, and clients. Whether it is worth the time and fee for you depends on your role and goals — it tends to pay off most for cloud engineers, architects, and consultants who work with AWS day-to-day or want to move into roles that do.
The passing score for AIP-C01 is 750 / 1000. The exam contains 75 questions and lasts 3 hr.
The AIP-C01 exam fee is $300 USD. Fees are set by AWS and may vary by region; always confirm the current price on the official AWS certification page before booking.
AWS certifications are valid for 3 years. Recertify by passing the current version of the same exam, or by passing a higher-level exam in the same path before expiration.
Yes. You can take the exam online (proctored via the provider's secure browser, available 24/7 in most regions) or at an in-person Pearson VUE test center during business hours. Both formats use the same questions, time limit, and passing score.
CertLabPro provides 15 study modes across the practice question bank for AIP-C01. The exam-simulation mode mirrors the real exam: 75 questions in 3 hr, with the same passing threshold of 750 / 1000. Browse mode lets you read every Q&A statically.