An HR manager in a hospital trust's administration team is reviewing staff guidance on choosing a Claude model. A colleague's draft says Haiku is the model for customer replies, Sonnet is the model for writing, and Opus is the model for numbers and spreadsheets. How should the HR manager correct the draft?
- AKeep the subject split but add Opus for policy work, since policy is the trust's most sensitive subject area.
- BReplace the subject split with a rule to use Opus for every task, so staff never get a weaker answer by mistake.
- CExplain that all three are general-purpose models, and staff should choose by task complexity, speed and cost. Correct
- DExplain that the models differ mainly in how private they are, so sensitive HR tasks go to the most private one.
Why A is wrong: Adding a rule for sensitive work feels prudent. But it keeps the false idea that each model is assigned to a subject, and sensitivity of content is handled by data policy, not by picking a model family.
Why B is wrong: A single rule is simple and seems to remove risk. But defaulting to the most capable model makes routine work slower and more costly for no benefit, which ignores the trade-off the families exist to offer.
Why C is correct: Haiku, Sonnet and Opus can each handle replies, writing and figures; what separates them is how much complex reasoning they bring against how fast and costly they are. Choosing by the demands of the task, not its subject, is the correct basis.
Why D is wrong: Linking sensitive HR data to a choice of tool sounds responsible. But the families are distinguished by capability, speed and cost; protecting sensitive data depends on the organisation's policy, approved accounts and removing identifiers, not on the model tier.