Pick model & effort
Build-along · ~8 min at your own keyboard
Two separate dials. Model: bigger for judgment and writing that matters, smaller for sorting and extraction. Effort (how long it thinks before answering): more for genuinely hard problems, less for easy ones. They cost differently: on a subscription you spend usage limits, on an API key you pay per token, and more thinking uses more of either. Where the effort dial lives changes between versions — at the time of writing it's in /model or set by asking it to think harder; check the docs. Test each dial on a fixed task and write down what you saw.
The long version: Choosing a model ↗
Do this
0/5Checkpoints
- ▸You should now see the model list after /model and know where effort is set.
- ▸You should now see four runs of the same task, each with a checklist score, time, and usage.
- ▸You should now see your routing rule written with real task names.
The three ways this goes wrong
- The model switch does not stick. — Run /model again and confirm the header shows the new model. Switching applies to new messages.
- The two results are not comparable. — Use the identical prompt and input in a fresh session for each.
- You judge on speed alone. — Judge quality and price for this task: could you ship the cheaper result?
Dated facts in this lesson: Claude Code lets you change model with /model and change reasoning effort or extended thinking separately; where the effort setting lives varies by version (researched Oct 2, 2026, not yet confirmed by a person: check the source before you rely on it).
Where these ideas come from: Effort (AI Hero dictionary). Paraphrased here.
Share what happened
Paste your checklist, the four runs with checklist score, time, and usage for each, and your routing rule.
Strip API keys, passwords, and tokens before you paste. Swap them for [KEY].
🔒 Don't paste confidential, personal-health, or payment info.