What is a reasoning model?
Updated · By Robert Breen
A reasoning model is an AI language model trained to work through a problem internally, step by step, before it gives its final answer. That extra thinking helps on tasks with several conditions or steps, at the cost of slower responses and more tokens used.
Why it matters for a small business
Many business decisions an AI might help with have more than one condition: a lead is hot only if the team is big enough and the timeline is short; an expense needs review if it fits two categories; a reply must answer three questions without promising anything outside the facts. Earlier models often fixed on one condition and missed another. Reasoning models check the conditions more carefully before committing.
The tradeoff is time and cost. The model's internal reasoning uses tokens you may pay for on the API, and answers take longer. For a short thank-you note, that is wasted effort. For sorting, checking rules or planning across several constraints, it is often worth it. Some providers let you set how much reasoning effort to use.
In a real lesson: Build an AI Demo Request Form That Qualifies Leads
In the AI demo request form lesson, the AI Agent rates every lead for Northwind Scheduling Software, a made-up software company. The system message sets the rules: "Hot: 5 or more staff and wants to switch within a month. Warm: a clear need but no timeline. Nurture: just looking." Hot needs two conditions at once, which is exactly where step-by-step checking pays off.
The model you pick is gpt-5-mini, which the voice-over calls fast, cheap and plenty smart for processing form submissions. OpenAI describes its GPT-5 models as reasoning models, so even this small one works through the rules before answering.
When you click Fill with test data and Book a Demo, the agent sees a team of twelve that wants to switch this month, rates it hot, saves a row to the Leads sheet and emails two demo times. Clear rules plus a model that checks them is what makes that rating dependable.
Try this lesson free or read the step-by-step guide.
Common confusions
Reasoning model vs chain-of-thought prompting
Chain-of-thought prompting asks any model to show its steps in the answer. A reasoning model does that kind of thinking on its own, usually without showing all of it. You do not need to ask a reasoning model to "think step by step".
Is a reasoning model always better?
No. For simple writing or quick lookups, a non-reasoning model can be faster and cheaper with equal results. Reasoning helps most when there are rules to apply, numbers to compare or several steps to plan.
Tips
- Write the decision rules out explicitly; a reasoning model applies clear rules well but cannot guess unstated ones.
- If responses are too slow or costly for a simple task, try a lower reasoning effort or a smaller model.
Related terms
Where you use it: free lessons
- Build an AI Demo Request Form That Qualifies Leads (Lovable and n8n, 15 min)
- Build an AI Agent That Categorizes Business Expenses (n8n, 12 min)
Frequently asked questions
- How do I know if a model is a reasoning model?
- The provider's documentation says so, often alongside a reasoning effort setting. Model names alone are not a reliable guide.
- Do reasoning models cost more?
- Often per task, because the internal thinking uses extra tokens on top of the visible answer. Check your provider's pricing for the specific model.