For multi-step problems, models often do better when they reason before answering.
Chain-of-Thought Prompting
Asking a model to work through a problem step by step — "think through the calculation before giving the final answer" — tends to improve accuracy on arithmetic, logic and multi-criteria decisions. Writing out intermediate steps gives the model more room to get each part right.
Reasoning Models
Many providers now offer models that are trained to reason internally before responding, sometimes with a configurable "thinking" budget. With these models:
- you usually don't need elaborate step-by-step instructions;
- a clear statement of the problem and the success criteria matters more;
- more reasoning costs more tokens and time, so match the budget to the task.
When Reasoning Helps
Maths, planning, code debugging, comparing options against several criteria, and questions requiring several pieces of information to be combined.
When It Doesn't
Simple lookups, classification, short rewrites and extraction rarely benefit, and reasoning adds latency and cost.
Practical Tips
- Separate the reasoning from the final answer, for example by asking for the answer in a clearly marked section or structured field.
- Check the reasoning when stakes are high, but remember that a plausible explanation doesn't guarantee a correct answer.
- Evaluate with and without extended reasoning on your own tasks to see whether it's worth the cost.