Choosing the right model

Matching model capability to the task — and knowing when premium pays.

Models come in tiers: frontier models with deep reasoning that cost real money, fast and cheap models that are astonishingly good at routine work, and purpose-built tools that beat both at one job. Using one model for everything — always the most expensive, or always the cheapest — is the tell of an operator who has not thought about it.

This guide teaches the judgment the exam calls "model–task fit": spending capability where stakes and complexity demand it, and economy where volume dominates. The judgment is durable even as specific models change monthly.

1. Stage 1 · E1 FOUNDATIONAL

   Learn that tiers exist

   Inside most AI products sits a choice of models: bigger, slower, smarter ones and smaller, faster, cheaper ones. The difference is real. Deep reasoning, subtle tradeoffs, and long complex documents favor the strongest models; reformatting, classifying, and tidying favor the fast ones — where the premium buys you nothing.

   Start noticing which you are using and what it costs (in money, time, or a monthly cap). If you have never switched models on purpose, switch once this week and compare.

   Try it this week

   Run the same routine task (say, cleaning up meeting notes) on a top-tier and a fast model. Then run a genuinely hard task (analyzing a tricky decision) on both. Note where the gap appears — and where it does not.

2. Stage 2 · E2 PROFICIENT

   Match capability to stakes and complexity

   Two questions size a task: how hard is the reasoning, and how expensive is an error? A board-level acquisition memo is hard reasoning and costly errors — use the most capable model and review carefully. Nine hundred rows of reformatting is easy reasoning and cheap errors — a fast model plus a spot-check wins on every axis.

   When a deadline tempts you to skimp on the high-stakes task, remember the asymmetry: compute is cheap next to a bad decision shipped to leadership. Skimping on the strategy memo is false economy; splurging on data cleanup is waste.

   Try it this week

   List today's AI tasks and label each: reasoning depth (low/high) and error cost (low/high). Assign a model tier to each quadrant, and see whether your actual habits match.

3. Stage 3 · E3 DISTINGUISHED

   Route work, escalate exceptions

   Volume work earns a pipeline: the cheap fast model handles the bulk, and the hard slice — the nuanced category it keeps misrouting — gets escalated to a stronger model or a human. That routing pattern (cheap by default, capable by exception) is how professionals get both quality and economy.

   Prefer purpose-built tools where they exist: a dedicated transcription tool with speaker labels beats a general assistant imitating one. The general model's job is the judgment work around the structured job, not the structured job itself.

   Try it this week

   Find one high-volume task where you use a premium model out of habit. Move it to a cheaper tier with a defined spot-check, and write the escalation rule for the cases the cheap tier gets wrong.

4. Stage 4 · E4 EXCEPTIONAL

   Reason about the portfolio

   The exceptional operator thinks in portfolios: across all their AI work, capability spend concentrates where judgment and stakes concentrate. They can defend every assignment out loud — "strongest model here because an error costs a client; fast model there because I verify a sample anyway" — in tier terms, without leaning on brand names.

   They also re-evaluate as models improve: yesterday's frontier task is today's mid-tier task. The judgment framework is permanent; the assignments are scheduled for review.

   Try it this week

   Write your model-assignment table: your recurring tasks, the tier each gets, the one-line reason, and the check that catches each tier's failures. This exact artifact is what a top exam answer looks like.

Ready to see where you stand? The free check scores all six competencies in about fifteen minutes.