Model comparisons

We run new models against the ones our own apps use, with the same inputs and the same prompt, and publish the real outputs, the time each run took and what it cost. Each page says which model we moved to and why.