effort means more than just "thinking time." Effort level controls how much work Claude does on your request overall. This does include how long the model thinks, but also:
How many files it reads;
How much it verifies; and
How far it pushes through a multi-step task before checking in with you.
At a higher effort, Claude will take more of those actions (for example, read files, run tests, and double-check) before it comes back to you. At lower effort, it would rather ask you for more context than spend tokens figuring something out on its own
They recommend starting with the default effort, as it fits most users and their tasks in terms of efficiency and token usage. Then if anything goes wrong based on the result switch the model, effort, or add more context.
The model setting is roughly how capable; the effort setting is roughly how thorough. Most real tasks need some of both.
https://claude.com/blog/claude-model-and-effort-level-in-claude-code
#llm
