I am currently learning about generative AI and working on integrating Gemini 2.5 Flash into my platform via API calls. Before proceeding with implementation, I want to clarify my understanding of the document-related technical specifications in the official documentation:
https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/2-5-flash
From the documentation, it states that the maximum file size per file is 50 MB and that the supported MIME types for document input are application/pdf and text/plain.
Based on this, I want to confirm whether my interpretation is correct:
-
Does this mean that, when using the API, only files corresponding to these MIME types (e.g.,
.pdf,.txt) are accepted as document inputs, and that formats such as.doc,.docx,.xls,.xlsx, or.csvare not supported directly? -
If those formats are not supported, is converting them into a supported format (for example, exporting to PDF or plain text) currently the only workaround?
-
Given the 50 MB per file limit, if a document exceeds this size (e.g., a 160 MB file), is the expected approach to split the document into multiple chunks, each under 50 MB, before submitting them to the API?
-
When using gemini google. com via the web UI, larger files and additional file types appear to be uploadable (potentially up to ~100 MB). Is this due to separate backend handling for the GUI, or are there supported API-side mechanisms that replicate this behavior?
-
Has anyone successfully uploaded larger files or non-PDF/text document formats through the Gemini 2.5 Flash API, or identified alternative patterns for handling these cases?
Any clarification or implementation insights would be greatly appreciated. Thanks.