- use ai v7 usage detail fields for reasoning and cache token counts. - keep context usage cost rendering compatible with the updated SDK types.
