GLOSSARY
Chain-of-Thought (Reasoning)
Letting a model work through a problem step by step before answering — dramatically better on math, logic and planning, at the cost of extra time and tokens.
Ask a model a hard question directly and it must commit to an answer immediately; ask it to “think step by step” and accuracy climbs — writing out intermediate steps lets each step condition the next, like showing your work in math class. Modern “reasoning models” (DeepSeek-R1, OpenAI's o-series) are trained to do this natively, producing long private deliberation before the visible answer.
The tradeoff is latency and cost: a reasoning model may burn thousands of tokens thinking about a question a standard model answers instantly. Practical rule of thumb — reasoning models for hard analysis, planning and debugging; standard models for chat, extraction and routine generation.