When to Use a Reasoning Model, and How to Prompt It
Reasoning models are better at hard problems, but they are slower and cost more. Using one for everything wastes time and usage limits. Using a fast model for a genuinely hard problem gets you a confident, wrong answer.
This lesson gives you a simple way to decide which to use, explains the thinking-effort settings many tools now offer, and shows how to prompt a reasoning model differently from a fast one.
What You'll Learn
- Which tasks benefit from a reasoning model, and which do not
- How to use thinking-effort or thinking-budget settings
- Prompting habits that work better with reasoning models
- A simple workflow that combines fast and reasoning models
Which tasks benefit
Reasoning helps most when a problem has several steps that depend on each other and a clear way to be right or wrong.
Hard, checkable, multi-step problems gain the most from extra thinking.
| Criteria | Use a reasoning model | A fast model is usually enough |
|---|---|---|
| Math and numbers | Multi-step calculations, word problems, budgets | Simple arithmetic, unit conversions |
| Logic and planning | Schedules with many constraints, puzzles | A simple to-do list |
| Code | Hard bugs, design decisions, tricky logic | Small snippets, formatting, explanations |
| Analysis | Comparing options against many criteria | Summarizing one document |
| Writing | Arguments that must hold together tightly | Emails, posts, rewrites, brainstorming |
Use a reasoning model
- Math and numbers
- Multi-step calculations, word problems, budgets
- Logic and planning
- Schedules with many constraints, puzzles
- Code
- Hard bugs, design decisions, tricky logic
- Analysis
- Comparing options against many criteria
- Writing
- Arguments that must hold together tightly
A fast model is usually enough
- Math and numbers
- Simple arithmetic, unit conversions
- Logic and planning
- A simple to-do list
- Code
- Small snippets, formatting, explanations
- Analysis
- Summarizing one document
- Writing
- Emails, posts, rewrites, brainstorming
A quick test: would a careful person need scratch paper for this? If yes, a reasoning model will likely help. If a person could answer off the top of their head, a fast model is probably fine.
Also consider the cost of a wrong answer. If a mistake would be expensive, like a financial calculation or a decision you will act on, the extra thinking time is cheap insurance.
Thinking-effort settings
Many tools let you control how much the model thinks, with names like reasoning effort, thinking budget, or modes such as "fast," "think," or "think longer." These are all dials for the same thing: how many thinking tokens the model may use before answering.
Decision
How much thinking should you allow?
- If Simple question or everyday writing
Fast model or low effort
Quick and cheap
- If A few connected steps
Medium effort
Good default for most hard tasks
- If Hard, high-stakes, many constraints
High effort
Slower, but more thorough
- If High effort and still wrong
Change the prompt, not just the dial
Missing information is the usual cause
More thinking is not automatically better. On easy tasks, extra thinking can make answers slower and sometimes worse, as lesson 5 explains. Start at a medium setting and raise it only when answers fall short.
How to prompt a reasoning model
Reasoning models do their own step-by-step work, so the most useful thing you can give them is a clear picture of the problem, not instructions on how to think.
Do:
- State the goal clearly. What exactly should the final answer be?
- Give all the constraints. Budget limits, deadlines, rules, things that must or must not happen.
- Provide the facts. Reasoning cannot make up for missing information. It can only work with what it has.
- Say how to check success. "The schedule must have no overlaps and every person must get at least one morning shift."
- Ask for the format you want for the final answer.
Avoid:
- "Think step by step." It already does.
- Dictating a rigid method, unless you really need that method. It can stop the model from finding a better route.
- Stuffing the prompt with loosely related material. Extra noise gives it more to reason about wrongly.
Example prompt:
I need a weekly shift schedule for 6 staff across 7 days, with 2 people on each morning shift and 2 on each evening shift. Rules: nobody works more than 5 shifts, nobody works evening then the next morning, and Ana cannot work weekends. Give the schedule as a table, then list each rule and confirm it is met.
The last line matters. Asking the model to check the result against each rule makes use of its strength: careful verification.
Combine fast and reasoning models
You do not have to pick one for a whole project. A practical pattern:
- Fast model to brainstorm, gather information, and draft.
- Reasoning model for the hard part: the calculation, the plan, the tricky decision, or checking the draft for errors.
- Fast model again to polish wording and format.
This keeps costs and waiting time low while putting the extra thinking where it pays off.
If you use Claude, Prompt Engineering for Claude covers its extended thinking settings in detail.
Key Takeaways
- Use a reasoning model for multi-step problems with a clear right answer, and when a mistake would be costly.
- A quick test: would a careful person need scratch paper?
- Thinking-effort settings control how much the model thinks. Start at medium and raise it only when needed.
- Prompt reasoning models with a clear goal, all constraints, the facts, and a way to check success, not "think step by step."
- Combine models: fast to draft, reasoning for the hard part, fast to polish.