I prefer to evaluate LLM costs based on actual tasks they help me with
context:
I have a 'legacy' Android app – with legacy xml views, Volley, tons of Java;
then there is an iOS version of the same app, a blend of Swift UI and UIViewController's; a bunch of oss dependencies for network, ui widgets, sqlite, etc
because on this app I am solo, I can't maintain two codebases and Compose Multiplatform is a perfect fit;
I asked gpt-5.5 to:
- download CMP-based app template, that supports iOS and Android targets
- align iOS codebase to that template; keep all logic in iosApp/, keep shared/ empty
- move Android legacy codebase into the new folder structure, but not as androidApp/ root (as it required to be Kotlin-only), but a sibling one; still referencing shared/
- make it work - everything builds, shared/ gets embedded into both apps; some test shared UI views successfully render on both platforms
it took it maybe 10-15 minutes, and the API cost was ~30-35 USD; of course, some steering was involved too;
I can complete the same task in a day, maybe bit faster; even if that takes me just 1 hour, an hour of my time costs a few times more than $35
so, it's a win, even with API pricing