reasoning model

What are Reasoning Models?

A reasoning model is a Large Language Models (LLMs) designed or trained to spend additional computation on a problem by generating a sequence of intermediate reasoning steps before producing its final answer.

Instead of immediately answering:

Question → Final answer

a reasoning model behaves more like:

Question → analyze the problem → try possible approaches → perform calculations → check the result → final answer

How is it different from a regular LLM?

Reasoning models are generally still Transformer-based LLMs. They are not necessarily a completely different neural-network architecture.

The main difference is their training and inference behavior.

Regular LLM	Reasoning model
Often answers relatively quickly	Uses more tokens and computation before answering
Optimized primarily for next-token prediction and instruction following	Further optimized for multi-step problem solving
May provide shallow reasoning	More likely to construct long reasoning paths
Lower inference cost	Higher inference-time computation and memory usage
Good for ordinary language tasks	Particularly useful for math, coding, logic, and planning