What is the primary trade-off when selecting a model tier?	Balancing the required output quality against the associated API cost.
Which model tier should be used for high-volume, low-complexity tasks like classification or extraction?	Claude Haiku (fastest and lowest cost).
Which model tier is best suited for deep reasoning, complex analysis, and multi-step coding?	Claude Opus (most capable, highest cost).
What is the primary benefit of Claude Sonnet?	It offers a balanced combination of intelligence, speed, and cost for general production work.
What is the recommended first step before writing production code?	Setting up a simple evaluation using 20-30 representative examples.
What is the recommended order for testing model tiers during evaluation?	Start with Haiku, and only step up to Sonnet or Opus if the quality is insufficient.
When should the most capable model, Claude Fable, be reserved?	For the toughest challenges where the extra capability justifies the significantly higher cost.
What API property reports the input and output tokens used, which determines the bill?	`response.usage`
In a real-world application, how should different types of tasks be handled?	By routing different tasks to different models within the same endpoint.
Which model tier should be the default choice for most general production work?	Claude Sonnet.