Chain of Thought

Spend tokens thinking before answering

Rather than emitting an answer immediately, the model writes intermediate steps first. Each step lands in the context, so the next one can build on it.

It helps on problems with multiple dependent stages — arithmetic, constraint checking, anything where a wrong first move poisons everything after it. On simple lookups it changes nothing.

The reasoning is not hidden machinery. It is tokens: they cost money, take time and occupy the window, whether or not the interface shows them to you. Reasoning models make this the default, which is why their bills and their latency look so different from a chat model's.