Gemini API adds page and spreadsheet-range selection, with billing for selected content
Google's November 18 Gemini API update adds file slicing: applications can select pages, worksheet ranges or media intervals before their contents reach the model. Google says only the selected content is processed and billed.
The useful change is the control over what gets read. An application reviewing a financial workbook can request a particular tab and cell range, while an agent working through a manual can fetch a few pages at a time. The announcement also covers video and audio intervals. Office-format support already appeared in the November 5 release notes; today's entry adds the selection controls.
For spreadsheets, the document-processing guide exposes sheet_index and an A1-style visible_cells range. Applications can request evaluated values, formulas or a rendered view of the sheet. The default layers include values and formulas, so inspecting the calculation need not require reading a screenshot.
Page selection works with inclusive, one-based start and end numbers. Callers can choose extracted text, page images or both. That allows an initial text pass followed by images of the pages whose charts or layout need closer inspection.
The same controls work in function responses, where a tool returns a file to the model. Gemini receives metadata including page counts and worksheet names automatically, allowing it to request narrower ranges on later turns. Google's example defines a page-inspection tool, lets the model call it, then returns the selected pages through the Interactions API. The application still supplies that tool and handles its arguments.
Slicing changes what the model ingests, but the document guide says upload-size and page-count limits still apply to the whole file. Selecting three pages does not make an otherwise oversized upload acceptable.
Comments
No comments yet. Be the first.