reasoning model
A reasoning model is a large language model trained to work through intermediate reasoning steps before giving an answer.
A reasoning model is a language model trained not to answer immediately but to first generate intermediate reasoning, breaking a problem down and checking its work, before producing a final answer. By spending more computation at inference time, it improves accuracy on complex tasks such as math, coding and multi-step planning.
Because it generates more tokens, a reasoning model is usually slower and more expensive than a standard language model. It is therefore often paired with smaller, faster models that handle easy tasks, with only the hard cases routed to the reasoning model.
This entry is based on AIPOST articles and widely known facts. If something is wrong, please send us a correction request.
Articles covering this entry
…backend, testing, documentation research or code review, giving routine checks a small, fast model and reserving a stronger reasoning model for careful reviews.
OpenAI and Synopsys are developing GPT-Synopsys to bring reasoning models into chip design work.
Salesforce has announced Cora, a CRM reasoning model that draws on 28 years of CRM experience and about a decade of data and AI research.
For this kind of workflow, the division of labor is straightforward: a reasoning model can decide what the task requires, while Jev selects among identified int…