# How to Choose the Right Claude Model Family on the CCAO-F Exam Choose the right Claude model family, Haiku, Sonnet or Opus, on the Claude Certified Associate exam and at work, by checking who reviews the output before it reaches anyone. A small accuracy gap picks the lighter model in one question and Sonnet in the next, and here's the reason, with both questions worked. From the book: [Claude Certified Associate Foundations Practice Questions](https://youcanbuildthings.com/books/claude-associate-practice-questions.md) Web page: https://youcanbuildthings.com/articles/claude-haiku-vs-sonnet-vs-opus Hello builders, You can choose the right Claude model family, Haiku, Sonnet or Opus, on every model question of the Claude Certified Associate exam, and at work the same afternoon, once you ask one thing first: who checks the output before it reaches anyone? In one question from my book, the model that missed 1 label in 40 is the right pick, because a person checks every label before it goes anywhere. Here is how the exam tests Claude Haiku vs Sonnet vs Opus, with both questions worked and a third for you to try. Product and Model Selection is 12% of the exam in Anthropic's guide, about 7 of every 60 questions. One candidate who works mostly with the developer tools passed and said "several questions focused on Anthropic products like Projects," and model choice sits in the same domain. If most of your Claude use is one long chat, this is the domain most likely to surprise you. ## What Anthropic says each model is for Anthropic's own [guide to choosing a model](https://academy.claude.com/tutorials/choosing-the-right-claude-model) describes each one in a line: > Haiku is fast and lightweight. > Sonnet is the daily driver. > Opus is a large reasoning specialist. For Sonnet it adds, "If you're not sure which model to pick, start here." For Opus it says the model "uses more of your rate limit, so you want to reserve it for tasks that really need it," and lists "Problems where you've tested with Sonnet and it struggled." The same page also lists a fourth model, Fable, for the largest projects, while the exam guide sticks to three, so the questions use Haiku, Sonnet and Opus. So we start on Sonnet, we move to Haiku for high-volume simple work where the limits matter, and we save Opus for a reason we can name: Sonnet already fell short, or the reasoning is hard enough that depth beats speed. ## Two questions, two answers ![Human review changes which Claude model to choose: in the Chapter 1 question she reviews every label by hand, so the lighter model at 39 of 40 is the key and the default model at 40 of 40 is the runner-up; in Mock 2.2 drafts go straight to a customer with only a quick glance, so Sonnet at 50 of 50 is the key and Haiku at 46 of 50 is the runner-up, failing 8% of replies customers see](https://youcanbuildthings.com/images/articles/claude-haiku-vs-sonnet-vs-opus/human-review-changes-claude-model-choice.png) **The first one.** A support lead must sort 2,000 short emails into four fixed categories, "with most of her week's usage already spent." She tested two models on the same 40 emails: the lighter one matched her own labels on 39, the default one on 40. She will review every label by hand before anything is sent. Which choice fits? The key is the lighter model. Her own review will catch a gap of one email in 40, and the seven words "most of her week's usage already spent" tell you what she's short of. **The second one.** A support team tested two models on 50 customer replies. Haiku's drafts passed 46, and Sonnet's passed all 50. Each draft "goes straight to a customer with only a quick glance from an agent." Which model fits? (Choose one.) - **A.** Sonnet, since errors would reach customers and its drafts passed every test. - **B.** Haiku, since 46 of 50 is a high pass rate and it uses far less of the limits. - **C.** Opus, since replies that go to customers deserve the most capable model. - **D.** Haiku, with a rule that agents read each draft more carefully than before. The answer is A, because nobody checks these drafts properly before they reach a customer, so the cheaper model would send a failed reply 8% of the time. B is the runner-up: the first question's instinct carried into a stem with no review to catch the misses. C is the bigger-model reflex, since Sonnet already passes everything. D answers a question about the review when the stem asks for a model. ## The check I run on every model question When I wrote the 25 model and product drills for the book, the deciding phrase was nearly always a volume, a deadline, a test result or a plan, and the two questions above add who reviews the work. I read the stem for these, in this order: - Who reviews the output before it reaches a customer, a client or a decision - How many items, and how fast they're needed - Whether a smaller model already passed the same test - How much of the usage budget is left When the stem names a reviewer, that line usually decides it. When it names nobody, look for speed, volume or a test result instead. Want to try one with no reviewer in it? Here's Drill 3.9 from the book. A support team wants replies drafted within seconds while customers wait in live chat. The replies are short, and a test showed the lightest model's drafts pass the team's checks. Which model fits? (Choose one.) - **A.** Sonnet with extended thinking on, so each reply is reasoned more carefully. - **B.** Opus, since customers waiting in a live chat deserve the most capable model. - **C.** Haiku, since speed matters most and its drafts already pass the team's checks. - **D.** Whichever model the agent prefers, since quality is about the same across all. Pick a letter and write it down before you scroll. --- The answer is C, decided by "within seconds while customers wait." The team's checks already passed the light model, and speed is the one thing the live chat needs. B is the runner-up, the bigger-model reflex again, and here it costs the speed customers are waiting for. Now go build something this weekend! John Cook --- Get the book: Claude Certified Associate Foundations Practice Questions, $12.99 on Amazon: https://amzn.to/4jEfE3L