Dictionary term
Reasoning Model
Also called thinking model · extended thinking · adaptive thinking
A reasoning model generates internal reasoning tokens (its 'thinking') before writing a final answer, which helps it plan, use tools and solve multi-step problems. Providers expose a setting, such as effort or reasoning effort, that trades answer quality against tokens and time.
In plain English
A reasoning model thinks things through before it answers. It is slower and uses more tokens, but it handles harder problems better.
Analogy
Like an employee who drafts working notes before sending the final email, rather than replying off the top of their head.
Why it matters
Use reasoning for multi-step decisions, coding and agent work, and lower the effort (or pick a non-reasoning model) for simple, high-volume steps where speed and cost matter more. On Anthropic's current models thinking is adaptive, steered by effort; the older fixed-budget 'extended thinking' mode is legacy.
In practice
An agent that reconciles failed Stripe payments against CRM records runs at high effort, while the step that tags inbound messages as Sales or Support runs at low effort to keep cost and latency down.
Related terms
Sources
Browse all terms →Last verified