Reference · Glossary

Reasoning model

Last updated

An LLM configuration that spends **extra compute on stepwise thinking** before answering — better on hard logic, slower and pricier than fast chat models.

When to use

Multi-step planning, constraint puzzles, tasks where cheaper models fail evals.

When not to

Bulk classification, simple rewrites, latency-sensitive high-QPS paths.

Example

A migration planner that must respect budget + downtime windows — run on reasoning tier; headline variants — fast tier.