Advanced

Reasoning models

When to reach for a thinking model over a fast one.

Knowledge cards
Best practice
Verify a Reasoning Answer by Its Load-Bearing Assumption
A verification habit for reasoning-model output: instead of re-reading the whole chain, ask the model which single assumption or input the conclusion most depends on, then check only that.
L3 · Workflows
Concept
Escalate to a Reasoning Model Only When the Fast One Fails
A routing rule: default every task to the fast model and only escalate to a reasoning model when the fast answer is demonstrably wrong or shallow — not pre-emptively.
L2 · Context engineering
Concept
Let Reasoning Models Think Before They Answer
A mental model for prompting ChatGPT's reasoning models: give them a well-defined goal and the constraints, then get out of the way — they do their own step-by-step internally.
L3 · Workflows
Knowledge card
Use a Reasoning Model to Produce the Plan, Not Just the Answer
A division of labour: use the reasoning model once to decompose a hard problem into an explicit, ordered plan, then execute the individual steps with the cheaper fast model.
L3 · Workflows
ChatGPT reasoning models

When to think slow

Reasoning models work through a problem step by step before answering, trading speed for depth. The skill is knowing when that trade is worth it. Reach for one when the task has multiple steps, subtle logic, real maths, or a plan to construct — anywhere a fluent but wrong answer would be costly. For everyday drafting, quick lookups, and simple rewrites, they are overkill and just slower. Matching the model to the task, rather than always using the strongest or always the fastest, is what separates efficient use from wasteful.

Brief them to use the thinking

A reasoning model rewards a well-framed problem. Give it the goal, the constraints, and the criteria for a good answer, and let it reason — do not over-specify the steps, since finding the steps is what it is good at. For hard problems, ask it to lay out its reasoning or consider alternatives before concluding, so you can follow how it got there. The value is in the thinking; a lazy prompt wastes it just as surely as using the wrong model does.

Frequently asked
What are ChatGPT's reasoning models?

Models that work through a problem step by step before answering, trading speed for accuracy on multi-step logic, planning, and maths. They shine where a wrong answer would be costly.

When should I use a reasoning model?

For complex logic, careful planning, tricky maths, or debugging. Use the fast default for everyday writing and quick questions where speed matters more than deep reasoning.

How do I prompt a reasoning model well?

Give it the goal, constraints, and success criteria, and let it find the steps rather than dictating them. Asking it to show its reasoning helps you follow and check the conclusion.