An AI model is the engine that reads your request and writes the answer. There is not one model, there are several, and they are not all the same engine. Think of them like vehicles: a freight truck, a sedan, and a scooter all move things, but you would not take a freight truck to grab milk down the street.
Every model balances three things at once. Push one up and the others move:
🧠
Capability
how well it reasons through hard, messy problems
⚡
Speed
how fast the answer comes back
💵
Cost
how much each request costs to run
🚚 Big "flagship" models: smartest, slower, pricier🛵 Small "fast" models: quick, cheap, fine for routine work🎯 The skill is matching the model to the job
A flagship model (the largest, most capable one a provider sells) is brilliant at hard reasoning. A small model handles everyday tasks for a fraction of the time and money. In a minute you will pick the right one for real jobs yourself.
02 / 05 · The myth
"Just always use the biggest, newest model."
This is the single most common mistake, and it is an expensive one. Let's name it, then correct it.
⚠ The myth
"Bigger and newer is always better, so default to the flagship for everything."
✓ The reality
A flagship is the smartest model, but it is also the slowest and the most expensive. For routine work (a reminder text, a quick FAQ answer, a tidy-up of some notes) a small, fast model is often just as good, returns in a fraction of the time, and can cost roughly 10 to 30 times less per request (figures vary by provider and change over time). Sending a reminder text does not need the freight truck.
🛵 Small, fast model
Your everyday workhorse.
✓ Fast: answers come back near-instantly
✓ Cheap: pennies, so you can run it at volume
✓ Great for routine, well-defined tasks
– Can struggle on long, layered reasoning
🚚 Flagship model
Bring it out for the hard stuff.
✓ Smartest: handles nuance and messy problems
✓ Best for analysis, code, tricky judgment calls
– Slower: more thinking takes more time
– Pricier: overkill (and wasteful) for simple jobs
The key: "best model" is not a fixed thing, it depends on the job. The goal is not the smartest model, it is the right-sized one. A well-written prompt on a small model often beats a lazy prompt on a flagship, which is exactly why /learn/prompting pays off.
03 / 05 · Watch it work
Pick the model for the job.
Choose a task. The page recommends a tier and explains why. This runs entirely in your browser, nothing is sent anywhere. The logic is a few plain rules, not a real AI call.
Capability needed
Speed priority
Cost per request
💵 Cost figures are illustrative only, approximate, and prices change. They show the relative gap between tiers, not a quote. Treat them as a mental model, not a price list.
Picking a model is not a one-time decision. Keep these in mind and you will avoid both overpaying and under-delivering.
1. Start small, escalate only if needed. Try a fast model first. Move up to the flagship only when the answers actually fall short.
2. Match the model to the task, not the headline. The newest model in the news is not automatically right for sorting your inbox.
3. Mix and match. Real systems often use a small model for routine steps and call the flagship only for the hard one. You do not have to pick just one.
4. A good prompt can shrink the model you need. Clear instructions often let a cheaper model do the job. See /learn/prompting.
5. Specifics drift, so do not memorize them. Model names, prices, and sizes change often. Memorize the trade-off (capability vs speed vs cost), look up today's specifics when you need them.
Snapshot, as of writing (mid-2026) · the one spot with volatile specifics · check current docs before relying on these
Tier
What it is for
Relative cost*
Small / fast (e.g. "Haiku"-class)
routine, high-volume, well-defined tasks
lowest (baseline)
Mid (e.g. "Sonnet"-class)
everyday work that needs solid reasoning
several times the baseline
Flagship (e.g. "Opus"-class)
hardest reasoning, analysis, judgment
often 10 to 30 times the baseline
*Illustrative only, approximate, prices change. Class names are examples of a typical three-tier line-up and will shift over time. Look up current model names and prices from your provider before you commit.
05 / 05 · Done
You now understand model choice better than most people who use AI daily.
You know models trade capability against speed and cost, why the biggest one is not always right, and how to match the tier to the job. That one habit quietly saves real money.
We pick and combine models so you get quality without overpaying. A small model for the routine steps, the flagship only where it earns its keep, wired together so you never think about it. That is the kind of thing we build, end to end.