Reasoning models

Why some models “think” before answering, and when that's worth the wait.

What it is

Reasoning models spend extra computation working through a problem before answering: breaking it into steps, checking and backtracking. Most assistants now switch this on automatically or let you choose it.

Why it matters

For math, planning, code and multi-step analysis, reasoning mode is often dramatically better. For quick rewrites it's just slower.

Key ideas

• More thinking time usually means better answers on hard problems.
• It costs more time and, on paid APIs, more money.
• It still makes mistakes, so check its work.

Try it: Same puzzle, two modes

1. Pick a multi-step problem from a problem set you've already solved.
2. Ask it with reasoning off, then on.
3. Compare against your own answer, not just against each other.