Fast AI Models Now Match Last Year's Frontier, But Engineers Still Resist Using Them

At the 2025 AI Engineer World's Fair, workshops reflected a growing industry focus on evaluating and trusting AI outputs rather than just building with the latest models. Despite this shift, many engineers continue defaulting to the most powerful frontier models even for simple tasks, according to observations shared by a developer on DEV Community. The author argues this habit is unnecessary, noting that today's faster, cheaper models now perform comparably to what frontier models could do roughly six months ago. Examples cited include Sonnet 4.6 matching Opus 4.1, and Gemini Flash 3.5 competing with Gemini Pro 3.1, while costing a fraction of the price and delivering faster responses. The piece calls for a broader change in developer mindset, urging engineers to consider fast models for routine tasks instead of automatically reaching for the most capable option.