Reasoning Model
A type of large language model that generates an explicit chain of thought before responding, solving complex problems step by step.
What Are Reasoning Models?
A reasoning model is a type of LLM that internally generates an explicit Chain of Thought before producing an answer, solving complex problems such as mathematical proofs and code generation in a step-by-step manner.
While a standard LLM predicts the next token "intuitively," a reasoning model decomposes a problem and reaches a final answer through self-verification of intermediate steps. It can also be seen as a design that embodies the concept of Test-time Compute scaling.
Representative Models
Examples include OpenAI's o series, DeepSeek-R1, and Anthropic Claude's extended thinking mode. All of these tend to achieve higher accuracy the more "thinking time" they are given, and a notable feature is that users can control the trade-off between the number of reasoning tokens and accuracy.
Identifying the Right Use Cases
Reasoning models are not a universal solution. For tasks that require immediate responses—such as simple summarization or translation—standard LLMs are more advantageous in terms of both cost and speed. In the author's experience, using a standard model for internal knowledge Q&A and a reasoning model for contract risk analysis tends to offer better cost-effectiveness.
The first step to adopting reasoning models is to clarify in advance which tasks genuinely require "thinking time."
Related Terms
Chain of Thought
A prompting technique that improves accuracy on complex tasks by having the LLM explicitly generate intermediate reasoning steps.
Multi-step reasoning
Multi-step reasoning is a reasoning approach in which an LLM arrives at a final answer not through a single response generation, but by going through multiple intermediate steps, such as generating sub-questions, verifying partial answers, and retrieving additional information.
LLM (Large Language Model)
LLM (Large Language Model) is a general term for neural network models pre-trained on massive amounts of text data, containing billions to trillions of parameters, capable of understanding and generating natural language with high accuracy.
