Reasoning model
What is Reasoning model?
A language model trained to work through a problem in steps before it gives an answer, instead of replying instantly. That extra "thinking" step measurably improves results on maths, code and logic. It also costs more and runs slower, so it isn't the right pick for every task. The first widely used ones appeared in late 2024.
Why it matters
A standard model answers more or less instantly, in one pass. A reasoning model is trained to work through a problem in steps before it commits to an answer, and that extra work measurably improves results on maths, code and logic. For a business the value is in the hard tasks: multi-step analysis, tricky debugging, anything where a fast but shallow answer is worse than useless. The cost is real too. The extra steps make it slower and more expensive per answer, so it is the wrong tool for simple lookups or quick drafts. Knowing when a task justifies it is most of the skill.
In practice
A team routes a routine “what are our office hours” question to a fast, cheap model, and a “reconcile these two conflicting financial reports and explain the gap” task to a reasoning model. The second problem needs the model to hold several facts, work through them and show its steps. Widely used reasoning models are a recent arrival, and the pattern since has been to match the model to the difficulty of the job.
