Anthropic - Claude Certified Developer - Foundations Study Guide

5: Model Selection and Optimization

This chapter grounds the model layer of a Claude application: how tokens, context windows, sampling, and next-token generation produce non-deterministic output, and how the request-shape options (fast mode, adaptive thinking, effort levels) steer quality against latency and cost. It then moves from the SDK and transport mechanics that wrap the Messages API to the concrete selection criteria across the four model tiers, the breaking behavior changes that ship with new model releases, and the token-budgeting, usage-tracking, cost-modeling, and prompt-caching techniques that keep a production workload affordable.