Reasoning Model

Reasoning models are trained or configured to spend more inference-time compute on multi-step problems, often by producing intermediate work before a final answer.

They can help on math, code, and planning, at higher latency and cost. They still need tools and verification for facts.