The short answer
A reasoning model writes out a long working-out before it answers you. That working-out is more predicted text, not a different kind of mind. It helps genuinely on problems with steps, because each step gives the next one something to build on. It costs more and takes longer, and on simple questions it buys you nothing.
The thinking is text you usually do not see
When a reasoning model works on a problem, it first generates a long internal passage: attempts, dead ends, corrections. Then it writes the answer you actually receive.
That passage is produced the same way as everything else. It is not deliberation in any human sense. But it works, because having written out step four, the model has step four in front of it when predicting step five.
Most products hide it, or show you a tidied summary. You are billed for it either way, which is the main reason these models cost more.
ORBIT / THE NAVIGATOR
Nothing mystical is happening while the little dots animate. It is writing notes to itself, quickly, and you are paying by the note.
Where it genuinely helps
Anything with steps that depend on each other. Multi-stage arithmetic. Logic puzzles. Code that must actually run. Planning where a wrong early decision poisons everything after it.
The gain comes from catching its own mistakes mid-working. A standard model commits to its first path and carries on. A reasoning model can notice a contradiction and start again, because it has room to.
When the extra cost is worth it
- Worth it
- Several dependent steps; a checkable right answer; an expensive mistake.
- Rarely worth it
- Drafting, summarising, rewriting, brainstorming, simple questions.
- Actively annoying
- Anything you want back quickly, where a decent answer now beats a better one later.
The shape of the trade rather than a rule. Try both on your own work; the difference is usually obvious.
What it does not fix
It does not make the model truthful. A long working-out can be confidently wrong from the second line onward, and the extra length makes it read as more authoritative rather than less.
It cannot know things it never learned. Thinking harder about a fact you do not have produces a more elaborate guess, not a fact.
And the summary you are shown is not a transcript. It is a cleaned-up account, so it is not evidence of how the answer was reached.
What to actually do
- Use it when the steps matter and the answer can be checked.
- Do not use it for first drafts. You are paying for thinking you did not need.
- Check the conclusion, not the working-out. Long and tidy is not the same as right.
BEEP / THE MYTH-BUSTER
I can produce four hundred words of immaculate reasoning towards an answer I made up at the start. Length is not evidence. It never was.
See it for yourself
Reading about it only gets you so far. This one is written to show you the thing described above, rather than describe it again.
A train leaves at 09:40 and the journey takes two hours and thirty-five minutes. What time does it arrive? Show your working, then tell me which step you were least sure of.
Then paste it into whichever assistant you use — ChatGPT, Claude, Gemini or Copilot. Not sure which? Meet them all.
If you remember nothing else
- The thinking is generated text, produced the same way as any other answer.
- It helps on problems with dependent steps, because each step informs the next.
- You pay for the working-out even when it is hidden from you.
- It does not make the model truthful, and cannot supply knowledge it lacks.
- For drafting and summarising, a standard model is faster and cheaper.
Where this came from
- OpenAI, Reasoning modelschecked 26 September 2026
Anything here that turns out to be wrong is worth telling us about. Figures that move — prices, sizes, energy use — are kept in the registry with their dates, rather than written into the prose where they would quietly go stale.
Questions about “Reasoning models”Asked by readers, answered by us. Open to read them.
Open to load them.
Ask about this page and we will answer it here, where the next person can read it too.
Nobody else can answer you, and you cannot answer anybody else. This is not a forum — we would rather say that than run a room we cannot watch at three in the morning.
Your question goes up with the answer and without your name. Nothing identifies you, ever.
If we cannot answer it we will tell you why, and that one stays between us.