Every generation on Breezify runs on a real language model, and you get to choose which one. That choice matters more than it might seem, because the model you pick affects both how capable the result is and how many credits the generation costs.
Haiku 4.5 is the fastest and cheapest option, and it is a genuinely good default. For a simple tracker, a small dashboard, or a landing page, Haiku usually gets you a clean, working app on the first try, and you can iterate quickly since each refine costs very little.
Sonnet sits in the middle. It handles more moving parts well: an app with several connected views, real data validation, or a workflow with multiple states, the kind of thing where a mistake in one place quietly breaks another. If your first Haiku attempt needed a lot of correction, Sonnet is usually the right next step rather than more retries on the same model.
Opus is for the apps where reasoning is the hard part, not the typing. Complex business logic, an unusual data model, or a spec with a lot of edge cases benefit from the extra thought Opus puts into a generation before it writes a single line.
A practical rule that works for most people: start on Haiku, and only move up if the result needs more correction than you would want to do by hand. You are never locked into one model either; you can switch per generation, so the model powering your next refine does not have to be the one that built the first version.