loading
loading
A mode where the model decides for itself when and how much to reason before answering, instead of you setting a fixed thinking budget.
On modern Claude models you set thinking: {type: "adaptive"} and the model picks how much internal reasoning each request needs — short for a lookup, deep for a multi-step proof — and interleaves thinking between tool calls automatically. This replaced the old budget_tokens knob, which is now deprecated or rejected with a 400 on newer models. The raw chain of thought isn't returned; you get either a summary (display: "summarized") or empty thinking blocks (display: "omitted", the default), and you must echo thinking blocks back unchanged when continuing on the same model.
Plainly
Think of Adaptive Thinking as the brain part that guesses or decides. A mode where the model decides for itself when and how much to reason before answering, instead of you setting a fixed thinking budget.
In practice
Use it when model choice, prompts, latency, cost, or quality affect the product result. In practice, define the owner, input, output, and failure mode before you rely on it.