A language model trained to work through an extended internal chain of thought and spend more compute on harder problems before answering; stronger on multi-step tasks.
Reasoning Model is a language model trained to work through a problem with an extended internal chain of thought before it answers, and to spend more inference compute on harder problems. Examples include OpenAI's o-series, Google's Gemini thinking modes, and Claude's extended thinking.
Where a standard model maps a prompt to an answer in one pass, a reasoning model first generates a long private deliberation, exploring steps, checking itself, and revising, then produces a final response conditioned on that work. The deliberation is usually hidden or shown only in summary. The gain shows on multi-step tasks such as maths, code, analysis, and planning; the cost is more time and compute per answer.
For AI visibility, reasoning changes how answers are assembled. A reasoning model can break a question into several sub-queries, retrieve and weigh multiple sources, and reconcile them before committing, so appearing across the many angles it explores matters more than matching one phrasing. It leans harder on trustworthy sources because it has the budget to check them.
