Gemini · Research

Multimodal

Ask across images and files at once, read charts with page numbers, question long recordings by timestamp, and run Deep Research on your own sources.

Gemini multimodal research

Why send the original

A transcript loses what was on the screen, and a text export of a PDF loses the charts. Google's API docs say the model reads PDFs with native vision, including charts and tables, and that plain-text formats lose them. When a question depends on what something looks like, send the file itself. This lesson covers combining files, long recordings, Deep Research and pulling data out of charts.

App limits and API limits differ

Don't mix up the numbers you read online. In the app, most files can be 100 MB, and video and audio length depends on your plan. On the API, PDFs can be up to 50 MB or 1,000 pages, and the File API takes video up to 20 GB on the paid tier. Figures in third-party articles often blend the two, or come from a year ago.

Frequently asked
Can Gemini read a chart in a PDF?

Yes. Google's document docs say the model reads text, images, diagrams, charts and tables in PDFs. Ask it to label which values are printed and which are estimated from the chart, then check them against the page.

How long can a video be in Gemini?

In the app, up to 5 minutes in total, or 1 hour on Google AI Pro or Ultra. Audio can be up to 10 minutes, or 3 hours on those plans. On the API, a 1 million token model handles about 3 hours of video at low resolution.

Can I use Gemini Deep Research with my own files?

Yes. Attach them with Add Files, and use Sources to choose Gmail and Drive. Deselect Google Search to limit research to your own material.