Reasoning effort
How hard the model thinks before it answers, and when to turn it up.
Most models let you set a reasoning effort, which is how much the model thinks before it acts. It's a separate dial from which model you pick, and using it well is one of the easiest ways to get better results without always jumping to the biggest model.
The levels
Effort runs low, medium, high, xhigh, max. Not every model offers the whole range. The most capable ones go highest, while Auto runs at a single fixed setting.
- Low: answer quickly, with minimal deliberation. Best for simple, well-defined tasks.
- Medium: a balance, and a sensible default for everyday work.
- High and up: think hard before acting. For problems with traps, like subtle bugs, decisions with downstream consequences, and anything where a fast wrong answer costs more than a slow right one.
More effort means more time spent reasoning before the model responds. On hard problems that usually means better answers, for a bit more time and cost. On easy problems it's wasted, since the answer was never in doubt.
Matching effort to the task
A practical approach: default to medium, drop to low for the obviously mechanical stuff, and bump to high the moment you hit something the model keeps getting wrong or a step where a mistake is expensive to unwind.
Two dials, not one
A strong model on low effort can be faster and cheaper than a weaker one grinding on high, and sometimes it's better too. There's no single best setting, it depends on the task. When you're stuck, try a more capable model and more effort, often in that order.
Where to set it
Effort is set per message, next to the model picker, so you can change it turn by turn. Set a default in Settings → Models and override it whenever a step deserves more or less.