Reasoning Model
Working knowledgeCore AI Concepts
A reasoning model is a class of AI model that works through a problem step by step before producing an answer, rather than generating a response immediately. It is more reliable on complex, multi-step tasks such as legal analysis, financial modelling and technical diagnosis, where a fast but shallow answer is worse than a slower, considered one.
In practice
The trade-off is cost and latency: thinking takes longer and uses more compute per query. Route work by value, sending hard, high-stakes questions to a reasoning model and routine ones to a faster, cheaper model, rather than paying reasoning prices across every interaction.