Pick the model and the effort before you start

TL;DR
- The dearer model costs about twice as much. It earns that on hard planning, not on routine work.
- Effort is the real control for how hard the AI thinks. "Think harder" in the prompt is not.
- Pick both at the start of a piece of work, and keep them to the end.
Most AI tools now let you choose a model. Some also let you choose how much effort it puts in. It is easy to leave both on whatever they were last time.
That is a choice too. It is just one you did not make on purpose.
This guide covers the two settings I pick before I start any real piece of work, and why.
What the models cost
These are Anthropic's prices for its two top models, per million tokens. A token is a piece of a word. A page of text is several hundred tokens.
| Model | Reading (input) | Writing (output) |
|---|---|---|
| Claude Fable 5.2 | $8 | $40 |
| Claude Opus 5.5 | $4 | $20 |
Prices from Anthropic's pricing page, as of November 2026. They change. Check the page before you plan around them.
So the larger model costs twice as much, on both sides. If you use an AI through a monthly plan rather than paying per token, the same gap shows up as how fast you use up your allowance.
When the dearer model earns it
Twice the price is worth it when the work is hard to get right the first time. Planning a large piece of work. Finding the cause of a problem nobody understands. Writing something where the thinking is the whole job.
It is not worth it for routine work. Writing up a plan that is already decided. Reformatting a document. Filling in a template.
Cole Medin describes a split many people now use, in a video from November 2026: plan with the larger model, then build with the smaller one. When the larger model hands out smaller jobs to helpers, set those helpers to the cheaper model too.
I do something close to it. My setup starts every new conversation on Opus. I choose Fable on purpose, for the larger pieces of work.
What effort does
Effort is a setting that controls how much the AI thinks before it answers. Low effort is quick and cheap. High effort is slower, costs more, and goes deeper.
Anthropic is clear that this setting is the control, not your wording. Its best practices page lists "think hard", "think harder" and "be thorough" in the prompt among the things you can stop doing, and says: "Effort is the control for that" (Best practices for Claude Code, as of November 2026).
So if an old prompt of yours begs the AI to think carefully, take that line out. Set the effort instead.
Two facts help you pick a level.
Most of the gain comes early. Anthropic's guide to Fable says: "You get most of the capability gain when using high effort" (Prompting Claude Fable 5.2, as of November 2026). The levels above high cost more and add less.
The newer models do more at each level. Anthropic's model page says Opus 5.5 "at medium matches or exceeds Opus 5 at high" (Model configuration, as of November 2026).
Defaults differ by tool. In Claude Code the default effort is medium. Through Anthropic's developer platform, Fable 5.2 defaults to high. Check which one you are on.
Keep one model for one piece of work
Switching models in the middle of a long conversation has a hidden cost.
Claude keeps a cache of the conversation so far. It saves money and time on every message. That cache belongs to one model. When you switch, the new model has to read the whole conversation again at the full price (Prompt caching, as of November 2026).
So pick the model at the start. If the work changes enough to need a different one, start a new conversation with a short handover, rather than switching partway.
The short version
Before a real piece of work, decide two things.
Which model: the larger one for hard thinking, the smaller one for everything else.
Which effort: high for work that matters, lower for routine work.
Then keep both until the work is done.
