A reasoning model is one built to think before it speaks. Instead of producing an answer in a single pass, it works through the problem in steps — a "chain of thought" — checking its own logic along the way. The result is markedly better performance on tasks where a glib answer is usually a wrong one: multi-step math, careful analysis, code, and legal reasoning.
The trade is time and cost. That deliberate stepping takes more compute and more tokens, so reasoning models are slower and pricier per answer. Recent releases like Claude Opus 4.8 and GPT-5.5 increasingly expose this as a dial — let it think harder when the problem is hard, keep it quick when it is not.
Why it matters at your desk. For a lawyer or researcher, this is the difference between an AI that drafts and one you can lean on for analysis — Opus 4.8 was the first model to clear the legal "all-pass" bar, where every clause must be right, not just most. Inside tools like Claude Projects, the same upgrade shows up as more reliable handling of genuinely complex work.
What to watch for: more thinking is not always better. For simple lookups it is wasted time and money, and a visible chain of reasoning can still be confidently wrong — a tidy-looking argument is not proof the conclusion is sound. Match the effort to the stakes.