Claude thinking levels: common questions about effort and cost
aistrategy
Plain answers on what Claude's thinking level means, when to raise it, and how it changes your bill.
What is the thinking level?
It is also called effort. It controls how long Claude reasons before it answers. The model is the size of the brain; the thinking level is how long it works on the problem.
Does more thinking always give better answers?
No. Returns diminish. In the guide's Opus 5.5 numbers, medium costs about 70% of high and gives up only 2.5 points, while extra high costs 2.5 times as much as high for 1.4 more points.
Can more thinking fix missing knowledge?
No. Thinking helps with planning, using tools and checking work, but it cannot add facts the model does not have.
Which level should I start with?
Medium on Opus 5.5 is the recommended starting point and the default in the Claude apps.
When should I raise thinking instead of switching models?
Raise thinking first when the answer is close but flawed, or when the task needs careful planning. Switch models when raising thinking does not help. Start the larger model at a lower level.
Can a bigger model be cheaper?
Sometimes. On a hard coding test, Fable 5.2 at high thinking matched Opus 5.5 at max for roughly half the cost per task. Cost depends on tokens used, not only the price per token.
Is low thinking ever the right choice?
Yes. Low costs about a third of high. Running low first and redoing only the misses scored about 97% at 17 cents per task, versus 95% at 29 cents with everything on high.
How do I see what a task costs?
Ask whether your AI tool reports cost per task. If it does not, that is worth raising with the vendor. Our AI consulting services can help you set up cost tracking that fits your workflows.