TinkerTwo questions, one effort setting.
Thinking recorded
Run a question to see what the model wrote before it answered.
Chapter 5 of 7
Effort is a hint, not a counter
The team sets effort to high and expects every answer to think the same amount. The token counts say otherwise.
This chapter: What does an effort setting actually control, in today’s APIs?
You will learn
- Why the same effort gives a short think on an easy question and a long one on a hard question
- What the current APIs take: effort levels and adaptive thinking, not a fixed token budget
- Why a hard cap on output still matters
Words you’ll meet (3)
- Adaptive thinking
- With adaptive thinking, Claude weighs each request and decides whether to think and how much: a simple factual question may get no thinking at all, and a multistep problem gets deeper reasoning.
- Thinking budget
- A fixed number of tokens a model may think for. The older way to control thinking; effort replaces it on newer models.
- max_tokens
- The hard limit on everything a reply writes, thinking and answer together.
The picture on the left is live. Each step below asks you to do one thing in it; the story continues here as you go.
Step 1 of 3
The team sets effort to high for two questions, a lookup and a delivery date, and expects them to think the same amount. Before you run them:
- “What is the cutoff time for same-day shipping?”
- “A customer orders an in-stock dock door sensor at 3 p.m. Central on a Friday, with expedited delivery. What day does it arrive?”