Putting DeepThink to Work on Hard Problems
DeepThink is DeepSeek's reasoning mode, powered by its R1 model line. It is the feature that made DeepSeek famous, because it matches much more expensive reasoning models at a small fraction of the cost. In this lesson you will learn what reasoning mode is actually good at and how to use it on real problems in math, logic, planning, analysis, and decision-making.
What You'll Learn
- What "reasoning" or "chain-of-thought" really means
- The kinds of problems where DeepThink shines
- Practical prompts for math, planning, analysis, and decisions
- How to read the thinking and still verify the result
What Reasoning Mode Actually Does
A normal chat model answers by predicting a fluent response in one pass. That works beautifully for writing and simple facts, but it can stumble on problems that require several careful steps, because it commits to an answer before working through the details.
A reasoning model like R1 does something different: it generates a chain of intermediate thinking first, working the problem out in steps, and only then produces a final answer. In the DeepSeek app you can usually see this thinking laid out above the answer. Because the model reasons before committing, it catches more of its own mistakes and handles multi-step problems far better. The trade-off is time: DeepThink answers take longer, sometimes many seconds, and that is normal.
| Criteria | Normal chat | DeepThink (R1) |
|---|---|---|
| How it answers | One fast pass | Steps through the problem first |
| Strongest at | Writing, quick facts | Math, logic, planning, debugging |
| Speed | Fast | Slower, shows its work |
| Catches its own errors | Less often | More often |
Normal chat
- How it answers
- One fast pass
- Strongest at
- Writing, quick facts
- Speed
- Fast
- Catches its own errors
- Less often
DeepThink (R1)
- How it answers
- Steps through the problem first
- Strongest at
- Math, logic, planning, debugging
- Speed
- Slower, shows its work
- Catches its own errors
- More often
Where DeepThink Shines
Reach for DeepThink whenever a problem has real steps and a wrong answer would cost you. Strong use cases:
- Math and quantitative problems: percentages, unit conversions, interest and loan math, probability, and word problems.
- Logic and puzzles: scheduling under constraints, seating arrangements, "who did what" deductions.
- Multi-constraint planning: trips, project timelines, or menus that must satisfy several rules at once.
- Analysis and critique: finding holes in an argument, comparing options against criteria, checking whether a plan is internally consistent.
- Careful code debugging: tracing why logic produces the wrong output (covered more in Module 3).
Practical Prompts
A money problem
Turn on DeepThink, then ask: "I borrow $6,000 at 9 percent annual interest, compounded monthly, and pay it back over 3 years in equal monthly payments. What is my monthly payment and how much total interest do I pay? Show the steps."
A constraint puzzle
"Five coworkers, Ana, Ben, Cara, Dan, and Eve, sit in a row. Ana will not sit next to Ben. Cara must be in the middle. Dan sits at one end. How many valid seating orders are there? Work through it step by step and list them."
A planning task
"Plan a 4-day trip to a coastal city for two people on a $900 total budget excluding flights. We like food and museums, dislike early mornings, and want one relaxed day. Give a day-by-day plan, keep a running budget total, and flag anything that pushes us over."
A decision analysis
"I am deciding between a $28,000 car with $1,200 a year in expected maintenance and a $34,000 car with $600 a year. I keep cars about 8 years. Compare the total cost of ownership, state your assumptions, and tell me which is cheaper overall and by how much."
Checking an argument
"Here is my argument for why our team should switch tools. Point out any weak reasoning, missing evidence, or hidden assumptions before I present it." Then paste your argument.
Reading the Thinking and Verifying
The visible thinking is useful for two reasons. First, it lets you follow the logic and spot exactly where an answer went wrong, which is far more helpful than a bare wrong number. Second, it makes it easy to correct course: if you see a bad assumption in step two, you can reply "you assumed X, but actually Y, redo it."
That said, reasoning mode is not magic and can still make mistakes. Keep these habits:
- Sanity-check the final number. Does it pass a rough estimate? If the loan math says your monthly payment is larger than the entire loan, something is wrong.
- Check the assumptions the model states. Reasoning answers often name their assumptions; make sure they match your real situation.
- Combine with Search for factual inputs. DeepThink reasons well, but if it needs a current interest rate or this year's rules, turn Search on so it reasons over correct facts.
- Do not use it as a shortcut past a vague question. If you have not said what you actually want, better reasoning just produces a confident answer to the wrong question.
When Not to Use DeepThink
For quick writing, brainstorming, rephrasing, translation, and simple lookups, DeepThink is overkill. You will wait longer for a result that normal chat would have produced just as well. The skill is knowing which problems deserve the extra thinking, and after this lesson you should be able to tell at a glance.
Key Takeaways
- DeepThink uses DeepSeek's R1 reasoning model to work through problems step by step before answering.
- It shines on math, logic, multi-constraint planning, analysis, and careful debugging.
- The visible thinking helps you find and correct exactly where an answer went wrong.
- Verify final numbers, check stated assumptions, and use Search to feed it current facts.
- Skip DeepThink for quick writing and simple questions, where normal chat is faster and just as good.