List price vs. the real bill

A model's bill is its price per token times the tokens it uses. Move the sliders to see when the "cheap" model stops being cheap.

The budget model's list price is 80% lower

Compared with the premium model's price per token.

But it uses 6.9x the tokens to finish the job

Extra "thinking" tokens and extra back-and-forth steps all count.

Premium

Budget

Illustration only. The 80% and 38% figures come from the study's Gemini 3 Flash vs. GPT-5.4 comparison; the 6.9x token figure is simple arithmetic from those two numbers, not a number the study reports.