onrup

Model architecture

Reasoning model

What is reasoning model?

A reasoning model has been trained to produce explicit intermediate steps before its final answer. The visible working improves accuracy on multi-step problems and costs proportionally more tokens per response.

The improvement is real on maths, code and multi-hop questions, and largely absent on simple retrieval or classification — where it just makes responses longer and slower.

A reasoning-distilled base is a strong starting point when the target task needs step-by-step behaviour, because the behaviour is present before fine-tuning begins.

Related terms

All terms in the glossary →

Start with the free tier

A magic link creates your account, your tenant and your first API key. No card until you ask for compute.