What is Reasoning model?

A model tuned to 'think' through problems with extended internal steps, trading speed and cost for higher accuracy on hard tasks.

A reasoning model is an LLM trained to spend extra compute thinking before it answers. Instead of generating a reply immediately, it produces a long internal chain of reasoning (often hidden or summarized), explores approaches, checks its own work, and only then writes the final answer. Examples include OpenAI's o-series and GPT-5 thinking modes, Anthropic's Claude with extended thinking, and DeepSeek R1. Many APIs let you control the effort level, trading answer quality against speed and cost. Reasoning models score much higher on math, science, coding, and multi-step planning benchmarks than standard models of similar size. The downsides are real: responses can take much longer, and the thinking tokens are billed as output, so costs rise. For simple tasks a fast standard model is usually the better pick.

Example

Given a tricky scheduling puzzle with overlapping constraints, a standard model often answers instantly and gets it wrong, while a reasoning model spends 30 seconds working through cases and returns a correct schedule.

Why it matters

Match the model to the task. Pay the reasoning premium for hard problems like debugging, math, and analysis, and use a cheaper, faster standard model for routine writing and lookups. Browse the AI tools directory or the model leaderboard to put it into practice.