Loading prices
List prices hide the burn. Once a month we build one reference app on each tool and write down what it actually consumed.
A small full-stack app with auth, a database table, and a deployed URL. The spec version is stored with the run so a changed brief does not get compared to an older one.
The app is built from scratch on each tool in the table, on the plan a solo builder would actually buy. No hand-tuning after the fact to make a tool look cheaper.
Credits used, tokens used, dollar cost, minutes, and whether the result ran. Those columns already exist. They are empty until the first monthly pass.
No benchmark runs are published yet. The method is fixed so the first month has something to compare against.