Most big AI labs ship model families with tiered names — Flash, Mini, Turbo, Lite on one end; Pro, Max, Ultra on the other. The naming is marketing, but the tradeoff is real: smaller models use fewer parameters, run faster, and cost less per token. Larger models are slower and pricier but handle harder reasoning, longer context, and edge cases better.

The interesting part is that "cheaper" doesn't mean lower quality across the board. A Flash-tier model often outperforms a Pro-tier model on straightforward tasks — at a fraction of the cost. Google's Gemini 2.5 Flash can "work harder" on complex tasks by using more reasoning steps, which pushes its cost closer to Pro tier on difficult jobs.

The practical lesson: the tier you need depends on your task, not prestige. Developers who default to the biggest model pay 5–10x more for gains they often can't measure.