Reasoning Model Glossary
A reasoning model is a model, or model mode, optimized for harder multi-step problems by spending more inference-time computation on planning, checking, and search before answering.
The term became common after OpenAI’s o-series and DeepSeek’s R1 made the distinction visible to ordinary users. In ordinary use, it marks the difference between a fast conversational model and one better suited to coding, mathematics, planning, debugging, or careful synthesis.
The term should not be taken too literally. A reasoning model does not necessarily reason the way a human does, and its visible explanation may not be a transparent record of its internal process. What matters operationally is that the system has been trained or configured to perform better on tasks where intermediate structure, self-checking, search, and inference-time effort matter.
For operators, the practical question is routing. Use reasoning models where the work justifies the latency and cost: proof-like reasoning, hard debugging, strategy synthesis, complex planning, and tasks where a cheap wrong answer is expensive.
See also
Chain of Thought · Model Tiering · Hallucination · GPT · DeepSeek