Alongside the mode picker in the composer there is an effort control. Where the mode decides which model answers, effort decides how hard it works on each reply.
The four levels
- Light — fastest. Good for lookups, quick rewrites, and short factual questions.
- Medium — the default, and the right choice for most work.
- High — more deliberation before answering. Worth it for debugging, multi-step analysis, and anything where a wrong answer is expensive.
- Extra High — maximum depth, and noticeably slower. Reserve it for genuinely hard problems.
It is saved with the conversation
Effort is set per conversation, not per message and not globally. Raise it once in a thread and every subsequent reply in that thread uses the new level; start a new conversation and it returns to your default. That makes it easy to keep one deep working thread alongside quick throwaway ones.
Effort and modes
Effort is a separate axis from Fast, Thinking, and Pro. Higher effort makes any mode more deliberate, but it does not turn Fast into Pro — the underlying model still differs. If you need the strongest model, change the mode; if you need the current model to try harder, raise the effort.
Higher effort means longer response times and, in Pro, more tokens against your budget. If replies feel slower than expected, check whether effort is still turned up from an earlier question.