FOUNDATIONS
Reasoning: More thinking – better answers?
Many everyday tasks require little extra reasoning: rewriting an email, shortening a text or generating a first set of ideas. For those, a model can often produce a good result within seconds.
Things look different with difficult logic, programming code, multi-step maths problems or decisions involving many variables. Here, so-called reasoning models (or reasoning modes) can spend additional compute on working through the problem before producing the final answer.
The scratchpad in its head
You can think of the difference like a difficult exam question:
- Answer immediately or check first: For simple tasks, the obvious answer may be enough. For complex questions, it can help to break the problem into smaller steps before committing to a result.
- More opportunities for error checking: Additional compute can help the model catch logical breaks, contradictions or calculation errors that are easier to miss in a fast response.
- More depth does not automatically mean more truth: Even a model that spends a long time reasoning can start from a false premise. Reasoning can improve logical performance – but it does not replace fact-checking.
Important: The explanation a model shows you is not necessarily a complete record of what actually happened internally. Visible intermediate steps are therefore not, by themselves, proof that an answer is correct.
THE RUNNING EXAMPLE
Mara plans a trip
For her five-day city break, Mara wants to stay within budget, sleep somewhere quiet and reach the sights she has chosen mostly on foot.
A cheap, quiet hotel on the outskirts may seem like a good fit at first. But long journeys and extra transport costs can reduce the price advantage. It is therefore not enough to tick off each wish separately: price, location and travel time need to be considered together.
Additional reasoning can help the model recognise these relationships and trade-offs. Perhaps no hotel satisfies every wish. Mara then has to decide which compromise works best for her.
Reasoning can help weigh several constraints together. It does not guarantee an error-free answer, and it does not decide for Mara which priorities matter most.
Your leverage in everyday use
- Fast mode for language: Use standard modes for wording, tone, creativity, translation, summaries and simple knowledge questions.
- Reasoning mode for structure: Use additional thinking time for complex logic, analysis, programming code, calculations, planning and problems with many constraints.
- Ask explicitly for care: Even without a dedicated reasoning mode, you can signal that speed is not the goal:
“Analyse the problem carefully. Break it into sensible steps, test my assumptions, consider counterarguments and check your result before giving the final conclusion.”
The key point: More thinking is not an automatic quality button. But wherever intermediate steps, dependencies and error checking matter, additional reasoning time can make a decisive difference.