Chain-of-Thought (Reasoning Model)
A model that writes out intermediate reasoning steps before its final answer, which measurably improves accuracy on math, logic, and multi-step problems.
Instead of jumping straight from question to answer, a chain-of-thought model generates a scratchpad of intermediate steps first — breaking a problem down the way a person might work through it on paper — and only then produces a final answer. Forcing that intermediate step, whether through prompting or through training the model to do it by default, tends to catch errors that a direct jump to an answer would miss.
'Reasoning models' that think longer before answering have become one of the main axes labs compete on, alongside raw parameter count — and a big part of why comparing a challenger model to GPT, Gemini, or Claude head-to-head now means comparing not just what they know, but how long and how carefully each one is willing to think before answering.