ToolNavs Find Useful AI Tools
Submit Sign in
Back to AI information
Gemini API Can Now Read Office Files Directly: Process Only What You Need, and the Bill Shrinks Too

Gemini API Can Now Read Office Files Directly: Process Only What You Need, and the Bill Shrinks Too

AI information • Admin • • 18 views

On November 18, Google confirmed a developer-facing change in the Gemini API changelog: Microsoft Office files such as Word, PowerPoint, and Excel, along with OpenDocument files, can now be passed directly to the Gemini API without first converting them to PDF or plain text. File slicing arrives alongside it, letting developers have the model process only the parts of a file they need, and pay only for that content.

How fine can you slice: from page numbers and cells to a stretch of video

The slicing granularity is the most practical part of this update. Documents can be cut by page range, spreadsheets by worksheet tab or cell range, and video and audio by time interval. You can also pick by content layer: just the images from one page, just the extracted text, just the formulas in a table, or just a video's transcript. Slicing works not only in user prompts but also for documents, spreadsheets, videos, and audio returned in function responses. In other words, developers no longer have to feed an entire file to the model just to ask about one page of it.

Who benefits most: long documents and pay-as-you-go use

Contracts, earnings reports, slide decks, and meeting recordings will feel the difference most: where asking about a single figure used to mean paying tokens for the whole file, now only the relevant pages, rows, or minutes go in. Apps for document Q&A and retrieval-augmented generation (RAG) should see thinner token bills, and with less irrelevant content to read, the model is less likely to drift off-topic. In the same week, Gemini Tasks went free in six countries for everyday users, so Google is pushing Gemini adoption on both fronts at once.

To save money, you have to specify the range yourself

Slicing is processing on demand, not automatic savings: without a specified range, the model still processes the whole file and bills for the whole file. Developers need to state the page numbers, cells, or time interval explicitly in the prompt or function call, and the changelog points to getting-started guides for document, video, and audio understanding. This change is about how files get into the model, not about the model's own capabilities, so developers can use it without switching models.

Recommended Tools

More