What You Need to Know
Objective 3.2 taught what the tiers are. Every choice between them spends three currencies: cost, time and depth. Cost is the plan's usage allowance, not a per-use invoice. Speed is the wait for each response, felt most sharply across many runs. Quality is depth on demanding work, as objective 3.2 defines it: holding ambiguity, reconciling contradiction and carrying judgement safely. The skill is naming the currency a task needs, then paying only that.
A fish and chip shop owner uses Claude for two jobs. One is five hundred review replies that vary only in a name and a dish. The other is one response to the council after a hygiene inspection.
Three Questions Name the Currency
Ask three questions in order. What does being wrong cost here? How many times will this run? Who is waiting for the answer? The answers name the currency. Without them, the choice is a vibe rather than a decision with named requirements. For the shop's replies: a cheap error, five hundred runs, nobody waiting. For the hygiene response: a costly error, one run, the council waiting.
Cost Means Your Usage Allowance
The Help Center lists model choice, effort level, current conversation length and file attachment size among the factors that determine how quickly usage is consumed. The allowance is finite, so every task you overpay for draws on the tasks still to come. Replies on the heaviest model eat allowance the hygiene response needs.
Volume Multiplies Cost and Speed
Run once, the difference between tiers is seconds and a sliver of allowance. Run five hundred times, the same difference decides whether the job finishes today and whether the allowance survives the week. High-volume routine work, like the review replies, belongs on the lighter tier.
Volume work on a fast tier gets a sampled spot-check, because Domain 2's verification duty never switches off. It scales sensibly: the owner reads a sample of the replies. The pairing makes the choice professional rather than merely cheap.
Stakes Multiply Quality
Stakes work on the other axis. Some single outputs feed a consequential decision and need depth: a regulatory submission, a legally sensitive letter, a recommendation a board will act on. The seconds a lighter tier saves are worthless against the cost of a subtly wrong conclusion. "It is only one letter" is true and irrelevant, because one letter with consequences out-demands five hundred without. Quality requirements follow the cost of being wrong, not the prestige of the task. High-stakes output then goes to qualified human review, whatever model drafted it. The hygiene response gets the capable tier and a qualified checker.
Split Mixed Workloads
Real workloads mix shapes. Split them by their demands rather than run everything at one setting. The routine bulk goes on the fast tier with sampled review. The hard core goes on the capable tier with proper review. The model picker makes the split cheap: the model can change at any point in a conversation and applies from the next response.
Three wrong answers recur. Maximum quality everywhere pays the depth price on work that does not need it, and across volume the waste compounds. Minimum cost everywhere ships shallow output on the work whose errors are expensive. The escalation loop re-runs a disappointing output on ever-heavier models. It treats a vague brief as a model problem, but a vague brief fails at every tier. Fixing the brief is Domain 1's move, and no tier upgrade substitutes for it. If the hygiene brief never names its reader, every tier misses.
In the exam
Stems ask you to price a real task in the three currencies. The guide's sample item keys a faster, lower-cost model for straightforward, high-volume drafting. The usual trap is volume language leaking into a stakes decision.
Key Concept
Volume multiplies cost and speed; stakes multiply quality. When a stem mixes both shapes, the keyed answer almost always splits the workload. The bulk gets the fast tier plus sampled review. The piece that carries consequences gets the capable tier plus proper review.
Exam Traps
Justifying the fast tier on a high-stakes output because it is only one letter
Stakes, not volume, set the quality requirement. One output feeding a consequential decision out-demands hundreds of routine ones, and the seconds saved by the lighter tier are worthless against the cost of a subtly wrong conclusion.
Justifying the top tier on routine bulk because the work is important
Importance does not transfer from a project to every task inside it. Routine transformation at volume needs speed and cost kept low, with a sampled check; paying the depth price hundreds of times over buys nothing the task needs.
Escalating to a heavier model when the real defect is a vague brief
A vague brief fails at every tier, so the escalation loop spends time and usage relearning that. When the stem shows a prompt missing audience, constraints or source material, the keyed move is fixing the brief, which is Domain 1 territory, not upgrading the model.
Practice Scenario
A regional athletics association is preparing its annual awards night: five hundred personalised invitation letters that vary only in name, club and event, and one difficult letter to a club being suspended for a safeguarding failure, where tone, precision and legal-adjacent care all matter. Spending on the invitations must stay as low as possible, and nothing that would make the suspension letter safer may be skimped. How should the administrator run the two jobs?
Build Exercise
Price a Mixed Workload in Three Currencies
What you'll learn
- How to name the currency a task needs most: cost, time or depth
- How to run bulk work on a lighter tier with a sampled review
- How to route a high-stakes output to depth plus qualified human review
- How to catch prestige and uniform-habit decisions in your own week
- Pick a real week that contains both a bulk task with many similar outputs and a single high-stakes task
Why: The mixed workload is the shape this objective tests; a week with only one shape hides the split decision.
You should see: Two named tasks: one bulk, one consequential.
- For each task, write one line naming the currency it needs most, cost, time or depth, and why
Why: Naming the currency is the whole discipline; every wrong answer in this objective skips this step.
You should see: Two one-line justifications, each naming a single dominant currency.
- Assign a tier to each task and run the bulk task on the lighter tier, spot-checking a sample against your quality bar
Why: The sampled check is what separates professional cost control from shipping unreviewed output.
You should see: Bulk output that passes the sample, produced noticeably faster.
- For the high-stakes task, note who gives it qualified human review after the capable tier drafts it
Why: Depth on the draft does not discharge the review duty; consequential output goes to a qualified human whatever model produced it.
You should see: A named reviewer with the standing to catch a subtle wrong conclusion.
- Review the week: mark any decision made by prestige or by uniform habit rather than by the task demands, and write the corrected assignment
Why: The two uniform settings, maximum quality everywhere and minimum cost everywhere, are the recurring distractors; auditing your own week makes them visible in stems.
You should see: Each task marked demand-led, prestige-led or habit-led, with corrections written for the last two.
Sources
- Models overview — Anthropic
- Usage limit best practices — Anthropic
- Change the model, effort, and thinking settings — Anthropic
- Claude Certified Associate – Foundations Exam Guide — Anthropic