Loading…
Glossary term
A model trained or configured to spend extra effort thinking through steps before answering.
A reasoning model is trained, often with reinforcement learning on problems with checkable answers, to spend extra effort thinking before it replies. It may produce a long internal chain of thought, check its work and try alternatives.
Many let users set how much effort to spend, trading cost and delay for accuracy.
Reasoning models have pushed performance on math, science and coding, at the price of more compute per answer.