The model menu at the bottom of a new chat holds two choices: the model, and the effort level beside it. Most people leave both where they are, and that has a cost, because each message you send makes Claude read the whole chat again. The longer the chat, the bigger the read, and high effort makes every read cost more. A long chat on high effort is what eats a weekly limit fastest.
Staying on one model keeps the chat cached, so each new reply only reads the new part, and that is why a long chat stays cheap. Change model part way through and the app shows a warning: the old model keeps the cache, and the new one reads the whole chat again, on its own limit. So decide first. Quick questions get low effort in a new chat, and hard problems get high effort in a fresh chat with only the context they need.
The facts come from Anthropic’s help page on changing the model, effort and thinking settings, and from the app’s own “Switch model?” dialog, as seen on 13 October 2026. Settings and limits change, so check the page for the current wording. This is an independent explainer, not affiliated with Anthropic.