Every model, priced live.

Input and output prices per million tokens, read from the catalog every ten minutes. Each price is the rate the model is served at upstream, per million tokens.