What is a “thinking” model

The usual model works like an advanced autocompleter: it predicts the next word based on the previous ones and gives the answer right away. This is great for something simple - rewriting a letter, explaining a term. The reasoning model (“thinking”) first writes itself a draft reasoning: it breaks the problem into steps, checks the moves, discards dead ends - and only then answers. Analogy: a regular model answers without thinking, like in a quick quiz; the thinking person solves the problem on a piece of paper, like on a math test. Under the hood, it spends more “internal” words and time, but it pulls out multi-step logic where the usual one stumbles. Leverage: the more steps, conditions and pitfalls in a problem, the greater the gain from reasoning. Insider: a thinking model is not “smarter with facts”, it is more careful in the chain of conclusions; asking a one-clap question will only waste her time. A typical beginner mistake is to turn on the “heavy” model for everything and be surprised at the delays. The rule is simple: complex logic is thinking, simple text is ordinary.