The AWS Certified Generative AI Developer - Professional exam (code AIP-C01) checks that you can put foundation models (FMs) into real applications on AWS: pick and call the right model, feed it the right data, wrap it in safety and security controls, keep it fast and affordable, and prove that it works. It's a professional-level exam, aimed at people with a couple of years of building on AWS and about a year of hands-on generative AI work.
AWS's exam guide splits it into five domains. Domain 1, foundation model integration, data management and compliance, is 31% of the scored questions: choosing models, data pipelines, vector stores, retrieval and prompts. Domain 2, implementation and integration, is 26%: agents and tools, deployment, enterprise integration and APIs. Domain 3, AI safety, security and governance, is 20%. Domain 4, operational efficiency and optimization, is 12%: cost, speed and monitoring. Domain 5, testing, validation and troubleshooting, is 11%.
The guide says the exam has 65 scored questions plus 10 unscored ones you can't tell apart, so 75 in all. The AWS certification page lists 180 minutes and 300 USD, at a test center or online. Scores run from 100 to 1,000 and you pass at 750. Scoring is compensatory: you need the overall score, not a pass in every domain. A blank answer counts as wrong and there's no penalty for guessing, so never leave one empty.
There are two kinds of question. Multiple choice has one right answer and three wrong ones. Multiple response asks for two or more answers out of five or more, and you only get the mark if you pick all of them. The practice here is single choice, but treat every option as true or false on its own, the way a multiple response question needs.
Most questions are a short story: a company, what it needs, and one or two constraints. The constraint picks the answer. Read the last sentence first, then hunt for the words that decide it: "least operational overhead" (managed and serverless beats code you run), "most cost-effective", "lowest latency", "without changing application code", "must not leave the VPC", "auditable". Two options often both work; the right one meets the constraint best.
The exam doesn't ask you to train models from scratch, do advanced machine learning, or build data engineering pipelines; the guide lists those as out of scope. It asks what a developer does with models that already exist. Services such as Amazon Redshift, AWS Budgets and Amazon Forecast are on the guide's out-of-scope list, so an option built around one of them is usually there to distract you.