An AI model is a trained computational system that processes inputs to produce outputs from learned patterns. In Claude, you choose a model for the task; models in the same family offer different balances of capability, speed, and token consumption. section 1.4.2 section 2.3
Start with the work to be done. A lightweight model such as Haiku can handle straightforward questions, file reviews, and quick edits, though it may need more careful prompting. Sonnet is a day-to-day option for coding; Opus is suited to deeper reasoning over complex conversations or large codebases. You can switch models within a conversation as the task changes. section 1.4.2
A model’s training data has a cutoff, so it may lack recent information. A client can supply current details such as the date, and web search can supplement older knowledge. Provide relevant details about your organization, location, or tooling in the prompt or supporting inputs; otherwise, the model may ask for clarification or make incorrect assumptions. section 2.3
Choose a model for a code review
For a quick file review, start with a lightweight model and give it the file and the review goal. For an architectural review across a large codebase, consider a deeper-reasoning model and supply the relevant code and constraints. Review the response for missing context and incorrect assumptions, then clarify and iterate. section 1.4.2 section 2.3

