Skip to content

How I Passed the Claude Certified Developer Exam (CCDV-F) With 926/1000

← Back to the Blog Menu

I passed Anthropic’s Claude Certified Developer – Foundations exam (CCDV-F) in September 2026 with 926 out of 1000. The pass mark is 720.

This is what I did, what actually mattered, and what I would tell anyone starting now. I won’t share any exam questions. Everything here is about how to prepare, not what was asked.

  • The exam tests understanding, not memory. Know why each feature exists and when to use it.
  • Anthropic’s own courses and docs are what count. I tried other material; only the official content turned out to matter.
  • Most questions have four plausible answers. One word in the question usually decides which one is the best practice.
  • The only real memorising is the common API errors. Expect one or two questions at most.

About a month, spread out rather than continuous.

  • Two intense weeks. In one of them I put in around four hours a day. I worked through every course Anthropic recommends when you sign up for the exam on its Skilljar academy, plus the optional ones, and took practice exams along the way.
  • The rest of the month: daily use. I already use Claude for most of my day-to-day work, so I kept doing that, but deliberately. Every time I hit a situation the courses had covered, I applied what I’d learned. After a couple of weeks it stopped being something I had to remember and became how I work.

That second part matters more than it sounds. The exam is built around scenarios, and the fastest way to recognise a scenario is to have been in it.

I used Anthropic’s courses and documentation, several free practice exams, and other free guides.

Having taken the exam, I’d say only the official material mattered. The free practice exams were useful for getting used to answering under time pressure, but they gave me a false sense of how hard the real thing would be. More on that below.

This was the biggest surprise.

By my count, around 85% of the questions I got were what I’d call trick questions. They give you a hypothetical scenario and four answers, and almost always all four answers are correct in general. Only one is the best practice for that exact scenario.

The difference is often a single word in the question, such as “occasionally” or “never”. Read the question twice before you read the answers.

Most free practice exams don’t prepare you for this. They follow an easy pattern: the longest, most detailed answer is right, and at least one answer is obviously wrong. A typical one asks why you’d use a system prompt in your code and includes an answer like “because I felt like it”. In the real exam I did not see a single question with an answer I could throw out on common sense alone. You can rule answers out, but only by knowing what each option is for and whether it fits the scenario.

So if a practice exam feels easy, don’t trust it.

For every feature, setting and parameter in the courses, ask three questions:

  1. Why was it built? What problem was it solving?
  2. Why is it configurable? What changes when you change it?
  3. When is it the right choice, and when is a different approach better?

If you can answer those, the scenario questions become a matter of matching the scenario to the reason a feature exists.

The one thing to memorise: common API errors

Section titled “The one thing to memorise: common API errors”

I did get questions on API errors, for example what to do when a request comes back with a 429. You won’t be asked about obscure codes. Focus on the common ones the courses highlight:

CodeWhat it meansWhat to do
401Problem with your API key: malformed, revoked or expiredFix the key; retrying won’t help
403The key lacks permission for that resourceFix the permission; retrying won’t help
429You hit a rate limit or spend limitBack off and retry, honouring the retry-after header
500Unexpected error on Anthropic’s sideRetry with exponential backoff
529The API is temporarily overloadedRetry with exponential backoff

The full list is on Anthropic’s API errors page. If you’re unlucky this is two questions; more likely it’s one.

Once the courses were done, I built my own practice simulator with Claude, and it was more useful than any practice exam I found.

The certification page has a downloadable PDF that lists the exam’s topics and how the questions are split between them. I gave that PDF to Claude and asked it to simulate exam conditions: the same number of questions and the same split across topics.

The feature I used most was focusing on a single topic. When one area kept costing me points, I switched the simulator to that topic only, so every question drilled the weak spot until it stopped being one.

You can build your own the same way. Download the topic outline from the certification page, attach it to a Claude conversation, and start with something like this:

You are an exam simulator for the Claude Certified Developer – Foundations
exam. The attached PDF lists the exam topics and how questions are split
between them.
Rules:
- Ask one multiple-choice question at a time, with four options (A-D).
- Every question is a realistic scenario. All four options must be
plausible; only one is the best practice for that exact scenario.
- Make the deciding detail a single word or condition in the question
(for example "occasionally", "never", "at scale").
- Never make the correct answer the longest one by default, and never
include an option that is obviously wrong.
- After I answer, tell me if I was right, why the correct option is the
best practice, and why each other option is not best for this scenario.
- Track my score per topic. After every 10 questions, show the scores.
Modes:
- "full exam": follow the topic split in the PDF.
- "focus: <topic>": ask questions from that topic only.
Start in full exam mode.

Two cautions: a simulator is only as accurate as the material you give it, so check its explanations against the official courses when something looks off. And treat it as practice for reading scenarios carefully, not as a preview of the real questions.

Do the official courses first, all of them, including the optional ones. Then use Claude every day and apply each thing you learned on purpose. Leave practice exams for last, and use them to train your reading, not to judge whether you’re ready.

The mistake I’d avoid is memorising facts. The exam doesn’t reward knowing what a feature is called; it rewards knowing when it’s the right tool.