Gemini API accepts more Office files and richer tool results
TL;DR
On November 5, 2026 the Gemini API added PowerPoint 97-2003, OpenDocument presentations, and OpenDocument spreadsheets as inputs, plus page images for older Word files. Function results can now return Office files, audio, video, and Files API URIs.
- Status
- Live
- Released
- Plans
- Gemini API
- Source
- —
What changed
The November 5, 2026 Gemini API release notes add file types and a wider function response.
Inputs. The API accepts PowerPoint 97-2003 (PPT), OpenDocument Presentation (ODP), and OpenDocument Spreadsheet (ODS). Word 97-2003 (DOC) files now get page-image processing. DOC, PPT, and ODP are rendered as page images. For ODS, the model receives the data and formulas from the first sheet only. Processing for DOCX, DOTX, ODT, PPTX, XLSX, and XLS is unchanged.
Function results. Tools can return Office and OpenDocument files, audio, and video, not only images, PDFs, and plain text.
- On the GenerateContent API, function response parts also accept
fileData, so a tool can return a Files API URI instead of inline bytes.displayNameworks on bothinlineDataandfileData. Video parts acceptmediaProcessingofSTATICorAGENTIC. - On the Interactions API,
function_resultsteps acceptdocument,audio, andvideoblocks, inline or by URI, in addition to text and image. Video blocks are inv1betaandv1alphaand acceptprocessing.
Google points to its Office and OpenDocument files page and its multimodal function responses page for the per-format rules.
Why it matters
Old Office binaries and OpenDocument files no longer have to be converted before they can be attached. Know what the model actually sees. A PPT or ODP becomes page images, so the text layer is whatever the renderer captured. An ODS file contributes the first sheet only. A second sheet is invisible unless you pull it out yourself.
Tool authors can hand back a slide, a spreadsheet, or a clip the model can read on the next turn, including by URI. That removes a class of "the tool saved a file the model cannot see" bugs. Video responses need an explicit mediaProcessing or processing value. Google's notes name STATIC and AGENTIC on GenerateContent.
Frequently asked questions
Does ODS send every sheet?
No. Google says the model receives the data and formulas from the first sheet.
How are PPT, ODP, and DOC files presented to the model?
As page images. Newer DOCX, PPTX, XLSX, and the other formats Google lists as unchanged keep their existing processing.
Can a tool return a file URI instead of bytes?
Yes on GenerateContent, via fileData. Interactions function_result steps can also pass a URI for document, audio, and video blocks.