Claude Certified Developer - Foundations

Expect a multiple-choice exam, with some items allowing more than one correct selection, delivered in a proctored setting. Read each item's instruction line carefully, since multi-select questions usually state how many options to choose. Confirm the current delivery method with the official certification body before you schedule.

Those figures are set by the certification body and are updated periodically, so they are not stated here. Check the official certification website for the current question count, time limit, and appointment length before booking, and plan your pacing practice around those numbers.

Passing thresholds and fees are published by the official certifying organization and can change between exam versions. Look them up on the official site rather than relying on forum posts or third-party summaries, which are frequently out of date. The same applies to any retake or rescheduling fees.

Retake rules, waiting periods, and any associated fees are governed by the certification body's policy and vary over time. Check the official site for the current policy before you sit the exam. If you do need a retake, use the score report's domain breakdown to target the areas you scored weakest in.

It is a foundations-level credential, so it does not assume production-scale ML experience, but it does assume you have written real code against the API. Candidates who have only read tutorials tend to struggle with scenario questions about prompt design trade-offs and multi-step agent orchestration. Hands-on building is the strongest predictor of comfort.

Start with API fundamentals, since every other domain assumes fluency in request and response structure, roles, and token accounting. Then move to prompt design, followed by tool use and agentic workflows, which build on prompting. Finish with context management, evaluation, and safety, which are easier to absorb once you have a working application to reason about.

Three recurring failure modes: confusing system-level instructions with user-turn content, mishandling the multi-turn tool-use loop by not returning results correctly, and treating context limits as a fixed budget rather than something to actively manage. A fourth is neglecting evaluation entirely, which shows up in questions about measuring quality and detecting regressions.

You should understand how models differ in capability, context window, and cost characteristics well enough to choose appropriately for a described scenario. Exact per-token prices change frequently, so questions based on memorized figures would age badly. Focus on the selection reasoning rather than the current rate card.

No official prerequisite course requirement is stated, and self-study is a legitimate path. That said, structured material helps if you are new to the API. Whatever route you take, pair it with a personal project, because the exam tests judgment about design choices that only becomes intuitive after you have debugged a few of them yourself.

Re-read the API documentation sections on messages, tool use, and caching, then write out - from memory - how a tool-use round trip and a cached prompt request are structured. Do a timed practice run to calibrate pacing, and review the official content outline one more time to catch any domain you have been avoiding.