Is that input or output tokens or both? That number sounds quite extreme. Maybe they include input tokens from deep research? That could be tens of thousands of

Is that input or output tokens or both? That number sounds quite extreme. Maybe they include input tokens from deep research? That could be tens of thousands of input tokens into a cheap model per task, for example.